4 · Analytic curves and skeleta

176 NAG Tags

Curves make the previous constructions visible. Nested discs describe points, annuli carry real intervals, and reduction gives the directions issuing from a type II point. We prove the basic calculations and then state the model-theoretic results that globalize them. Throughout the curve classification and residual-curve discussion, KK is algebraically closed, complete and nontrivially valued. The explicit annulus calculations themselves work over an arbitrary complete non-Archimedean field.

4.1 Disc seminorms and the four kinds of points

For a∈Ka\in K and r≥0r\geq 0, define a seminorm on K[T]K[T] by ∣P∣a,r=max⁡j∣cj∣rj,P(T)=∑jcj(T−a)j,|P|_{a,r}=\max_j |c_j|r^j, \qquad P(T)=\sum_j c_j(T-a)^j, where for r=0r=0 only the constant coefficient is retained. Proposition 1.2.2 proves multiplicativity for r>0r>0; for r=0r=0 this is evaluation at aa. Write ζa,r\zeta_{a,r} for the resulting analytic point. If ∣a−b∣≤r|a-b|\leq r, translating the polynomial in both directions gives ∣P∣a,r=∣P∣b,r|P|_{a,r}=|P|_{b,r}. Thus the disc, not its chosen centre, determines the point.

Theorem 4.1.1. NAG-000160 Every point of AK1,an\mathbb A^{1,\mathrm{an}}_K is either ζa,r\zeta_{a,r} for some a,ra,r, or the limit of the seminorms of a decreasing sequence of closed discs with empty intersection. The four types are:

  1. Type I: r=0r=0, an evaluation at a∈Ka\in K.

  2. Type II: r>0r>0 and r∈∣K×∣r\in|K^\times|.

  3. Type III: r>0r>0 and r∉∣K×∣r\notin|K^\times|.

  4. Type IV: the empty-intersection case.

Proof. For a seminorm xx, set r=inf⁡a∈K∣T−a∣xr=\inf_{a\in K}|T-a|_x. Suppose the infimum is attained at aa. If ∣a−b∣>r|a-b|>r, the ultrametric inequality with unequal sizes gives ∣T−b∣x=∣a−b∣|T-b|_x=|a-b|. If ∣a−b∣≤r|a-b|\leq r, the same inequality gives an upper bound rr, while the definition of the infimum gives the reverse bound. Hence ∣T−b∣x=max⁡(r,∣a−b∣).|T-b|_x=\max(r,|a-b|). Every polynomial splits into linear factors over KK, so this identity and multiplicativity identify xx with ζa,r\zeta_{a,r}.

If the infimum is not attained, choose aia_i such that ri=∣T−ai∣xr_i=|T-a_i|_x decreases strictly to rr. For j>ij>i, one has ∣ai−aj∣=ri|a_i-a_j|=r_i, so the discs Di=D(ai,ri)D_i=D(a_i,r_i) are nested and ri∈∣K×∣r_i\in|K^\times|. A common point bb would give ∣T−b∣x≤ri|T-b|_x\leq r_i for every ii, attaining the infimum, a contradiction. The radii cannot tend to zero, since completeness would then give a common limit of the centres.

For each b∈Kb\in K, eventually ri<∣T−b∣xr_i<|T-b|_x, so ∣ai−b∣=∣T−b∣x|a_i-b|=|T-b|_x and ∣T−b∣ai,ri=∣T−b∣x|T-b|_{a_i,r_i}=|T-b|_x. Splitting a polynomial into its finitely many roots proves convergence to xx. Conversely, a nested family of discs with empty intersection gives decreasing nonnegative values on each polynomial; their limits preserve products and the ultrametric inequality. A nonzero polynomial has all its roots outside a sufficiently small member of the family, so its value is then a nonzero constant. The resulting limit is a norm and cannot be a disc seminorm, by the empty-intersection property. ◻

A field is spherically complete when every nested family of closed balls has nonempty intersection. Over a complete field, empty nested families can be reduced to a sequence by choosing radii approaching their infimum. Thus spherical completeness excludes type IV points. Ordinary completeness alone only excludes nested empty families whose radii tend to zero.

References. [TemIntro] (2.3.3); [CLD25] (11.1.13). The proof above uses algebraic closedness in the factorization of polynomials.

What changes over the original ground field

For a complete field K0K_0 which is not algebraically closed, form C=K‾0^C=\widehat{\overline K_0} and use the scalar-extension projection AC1,an→AK01,an\mathbb A_C^{1,\mathrm{an}}\to\mathbb A_{K_0}^{1,\mathrm{an}}. The classification above applies upstairs. A rigid point downstairs can have completed residue field K0[T]/(P)K_0[T]/(P) for an irreducible polynomial PP, and need not be a K0K_0-rational point. If PP is separable of degree dd, its base change to CC consists of dd classical points, one for each root. Algebraically this is the product decomposition C[T]/(P)≃CdC[T]/(P)\simeq C^d.

Thus the phrases “classical point” and “evaluation at an element of KK” coincide in the algebraically closed discussion, but not over a general ground field. Likewise residual directions over a non-algebraically-closed residue field carry degrees. These degrees must be restored when descending a geometric slope or measure formula. The completed-tensor-product construction in Section 2.6 gives the precise base change; a picture of discs over K0K_0 alone does not.

4.2 Residue fields and topology of the line

The point classification has a valued-field interpretation. For a type II point, translate and scale to the unit Gauss point; its residual field is K~(t)\widetilde K(t) and it adds no value-group rank. For a type III point, algebraic closedness makes ∣K×∣|K^\times| divisible, so its radius is rationally independent of the old value group. Proposition 3.8.1 gives the following table.

Type sK(x)s_K(x) tK(x)t_K(x) Description
I 0 0 H(x)=K\mathcal H(x)=K
II 1 0 H(x)~≃K~(t)\widetilde{\mathcal H(x)}\simeq\widetilde K(t)
III 0 1 v(H(x)×)=Γ+Zv(T−a)v(\mathcal H(x)^\times)=\Gamma+\mathbb Zv(T-a)
IV 0 0 immediate extension

For completeness, the last row can be checked directly. If xx is represented by empty nested discs, each rational function has, on a sufficiently small disc, the same norm as its value at the centre. Moreover, after dividing by that central value its residue is one: for a linear factor T−bT-b outside that disc, the ratio (T−b)/(ai−b)(T-b)/(a_i-b) differs from one by an element of norm less than one. Products and quotients give the assertion for rational functions. Completion changes neither values nor residues, as proved in Proposition 3.8.1. Thus the extension is immediate.

The topology is the evaluation topology, not the metric topology of KK extended by a few points. Increasing radii give continuous paths r↦ζa,rr\mapsto\zeta_{a,r}. Two discs join through the smallest disc containing both; for D(a,r)D(a,r) and D(b,s)D(b,s) its radius is R=max⁡(r,s,∣a−b∣)R=\max(r,s,|a-b|). Along a radial segment the natural length is a difference of logarithmic radii, so the distance between the two nonclassical disc points is ℓ(ζa,r,ζb,s)=log⁡(R/r)+log⁡(R/s).\ell(\zeta_{a,r},\zeta_{b,s})=\log(R/r)+\log(R/s). A classical point is approached as a radius tends to zero and is at infinite distance in this logarithmic metric. Type IV endpoints are obtained by the nested-disc limit.

Example 4.2.1. (The two topologies need not agree). At the unit Gauss point of P1,an\mathbb P^{1,\mathrm{an}}, the inward branches are ∣T−a∣<1|T-a|<1, indexed by a~∈K~\widetilde a\in\widetilde K, and there is one outward branch. A basic ordinary neighborhood is specified by finitely many polynomial evaluations, so it contains every inward branch except possibly finitely many. Indeed a polynomial has constant Gauss size on each branch not containing a residual root of its normalized reduction.

Choose points at a fixed positive logarithmic distance from the Gauss point in distinct inward branches. They converge to the Gauss point in the Berkovich topology but not in the path metric. The path metric is useful on a fixed finite skeleton, yet it must not replace the ambient topology in statements about compact support or weak convergence of currents.

4.3 Annuli and a retraction with an explicit homotopy

Let 0<r≤R0<r\leq R. The closed annulus A[r,R]A[r,R] has affinoid algebra K{R−1T,rT−1}={∑m∈ZamTm:∣am∣Rm→0(m→+∞),∣am∣rm→0(m→−∞)}.K\{R^{-1}T,rT^{-1}\} =\left\{\sum_{m\in\mathbb Z}a_mT^m: \begin{array}{ll} |a_m|R^m\to0& (m\to+\infty),\\ |a_m|r^m\to0& (m\to-\infty) \end{array}\right\}. Its spectral norm is the maximum of its two endpoint Gauss norms. The skeleton is the subset Σ={ζ0,ρ:r≤ρ≤R}\Sigma=\{\zeta_{0,\rho}:r\leq\rho\leq R\}. The function t=−log⁡∣T∣t=-\log|T| identifies it with [−log⁡R,−log⁡r][-\log R,-\log r].

An annular skeleton with attached branches and its logarithmic interval
The highlighted Gauss points form the annular skeleton. The logarithmic coordinate maps it to the interval below and contracts the attached branches; only a few branches are drawn.

Proposition 4.3.1. NAG-000150 The map τ(x)=ζ0,∣T(x)∣\tau(x)=\zeta_{0,|T(x)|} is a continuous retraction onto Σ\Sigma. There is a strong deformation retraction from the identity to τ\tau which preserves ∣T∣|T|.

Proof. Write P(X+Z)=∑jPj(X)ZjP(X+Z)=\sum_jP_j(X)Z^j for a polynomial, using its Taylor coefficients without division by factorials. For u∈[0,1]u\in[0,1], define a seminorm by ∣P∣H(x,u)=max⁡j∣Pj(x)∣ (u∣T(x)∣)j.|P|_{H(x,u)}=\max_j |P_j(x)|\,(u|T(x)|)^j. This is evaluation at T(x)+ZT(x)+Z followed by the Gauss norm of radius u∣T(x)∣u|T(x)| over H(x)\mathcal H(x), hence multiplicative. Its value on TT is ∣T(x)∣|T(x)|, so it defines a point of the same annulus. For u=0u=0 it is xx. For u=1u=1, translation by the scalar T(x)T(x) preserves the radius-∣T(x)∣|T(x)| Gauss norm, since this scalar has exactly that size; hence the point is ζ0,∣T(x)∣\zeta_{0,|T(x)|}.

For a skeletal point x=ζ0,ρx=\zeta_{0,\rho}, each Taylor term has size at most ∣P∣x|P|_x, while the j=0j=0 term already has that size. Thus the homotopy fixes Σ\Sigma. Finally the displayed formula is continuous in (x,u)(x,u) for every polynomial. Inversion of TT is continuous on the annulus, and Laurent polynomial approximations converge uniformly in its affinoid norm. Therefore all analytic evaluations are continuous along the homotopy. This proves the strong deformation retraction. ◻

The same formula restricts to open annuli because it preserves ∣T∣|T|. This proof also explains what a retraction forgets: the seminorms of individual functions can change along a fibre of τ\tau, even though the radius is fixed. A function factors through τ\tau only when it is constant on those fibres.

4.4 Units on annuli and integral slopes

Proposition 4.4.1. NAG-000152 For a nonzero analytic function f=∑amTmf=\sum a_mT^m on a closed annulus, its skeletal function is F(t)=−log⁡∣f(ζ0,e−t)∣=min⁡m{v(am)+mt}.F(t)=-\log|f(\zeta_{0,e^{-t}})|=\min_m\{v(a_m)+mt\}. It is piecewise affine with integral slopes. If ff is an analytic unit on the annulus, FF is affine.

Proof. The formula is the Gauss norm. Uniform convergence of the Laurent series and the positive minimum of its Gauss norm on the compact radius interval show that only finitely many terms can attain the minimum. Thus FF is the minimum of finitely many affine functions with integral slopes. In logarithmic radius s=log⁡ρs=\log\rho, the function log⁡∣f∣\log|f| is convex. If ff has an inverse, multiplicativity gives log⁡∣f−1∣=−log⁡∣f∣\log|f^{-1}|=-\log|f|, and this negative is convex as well. Therefore log⁡∣f∣\log|f|, and hence FF, is affine. ◻

Example 4.4.2. For f=T−bf=T-b with r<∣b∣<Rr<|b|<R, one has F(t)=min⁡(t,v(b))F(t)=\min(t,v(b)). The slope drops by one when the radial path passes the radius of the zero. On the full analytic curve, a branch toward bb is attached there; along that branch −log⁡∣T−b∣-\log|T-b| increases with slope one. The additional branch restores the zero total outgoing slope at the nonclassical attachment point. Looking only at the interval would lose this contribution.

A monomial map T↦cTmT\mapsto cT^m sends the skeletal coordinate to mt+v(c)mt+v(c) and multiplies lengths by ∣m∣|m|. Its injectivity on an interval does not make it unimodular: the tangent-lattice index is ∣m∣|m|. This distinction will reappear in tropical multiplicities and faithful tropicalization.

4.5 Residual curves and tangent directions

Let XX be a smooth proper connected algebraic curve over KK. At a type II point xx, the field H(x)~\widetilde{\mathcal H(x)} is a one-variable function field over K~\widetilde K. Let CxC_x be its smooth projective curve. A tangent direction at xx is a germ of a nontrivial path leaving xx; equivalently it is a branch in a sufficiently small suitable neighborhood.

Theorem 4.5.1. There is a canonical bijection TxXan≃Cx(K~)T_xX^{\mathrm{an}}\simeq C_x(\widetilde K) at a type II point. A type III point has two directions, and a type IV point has one. In a semistable model containing xx as a vertex, CxC_x is the smooth normalization of the residual component attached to xx; nodes determine distinguished directions.

Proof. Write k=K~k=\widetilde K. By [StacksAlg] (Tag 0A26) the proper curve is projective. The empty-marking case of Theorem 4.38.2 gives a projective split nodal model with the specified generic fibre. By Theorem 4.13.2 the analytic generic fibre of its completion is XanX^{\mathrm{an}}. Given a type II point xx, apply Theorem 4.27.4 to {x}\{x\}. On the resulting model Theorem 4.29.2 identifies the normalization of the component at xx by its function field H(x)~\widetilde{\mathcal H(x)}, and identifies its closed points with the intrinsic branches at xx, independently of the model. The normal proper curve with this function field is its nonsingular projective model, which is smooth because kk is perfect [StacksAlg] (Tag 0BY3). Its closed points are kk-rational because kk is algebraically closed. On a model containing xx the cofinal punctured stars of Theorem 4.25.3 identify the two ends of a self-node with distinct normalization points.

For the remaining valences use the open disc and open annulus tubes of Theorem 4.23.3. Their complement consists of type II component points. Each tube is an open subspace of an analytic line in its chart, so it suffices to count local branches there. Use the disc classification of Theorem 4.1.1. At a type III point x=ζa,rx=\zeta_{a,r}, r∉∣K×∣r\notin|K^\times|. The circle ∣T−a∣=r|T-a|=r consists of xx: the nonzero monomials in each polynomial in T−aT-a have distinct norms, since ∣K×∣|K^\times| is divisible. Open annuli with radii straddling rr are cofinal neighbourhoods. Indeed finitely many polynomial evaluations determine a basic neighbourhood, and after factoring those polynomials the absolute values of their linear factors near this circle are either constant or equal to ∣T−a∣|T-a|. Each punctured annulus has precisely two components, its inner and outer open annuli. Each is path connected by the radial homotopy of Proposition 4.3.1; shrinking preserves the two labels.

At a type IV point take nested closed discs with empty intersection representing xx, passing to a subsequence with strictly decreasing radii. Writing the discs as D(an,rn)D(a_n,r_n), use the open discs Un={∣T−an+1∣<rn}U_n=\{|T-a_{n+1}|<r_n\}. Each UnU_n contains D(an+1,rn+1)D(a_{n+1},r_{n+1}) and lies inside D(an,rn)D(a_n,r_n). Thus these neighbourhoods of xx are cofinal: eventually the discs exclude all the roots of any given finite list of nonzero polynomials, so those polynomial norms are constant on the disc. Each such open disc remains path connected on deleting xx. Indeed any two other points in it can be enlarged to the same closed disc strictly inside it, centred at its centre, with radius larger than both values of the centred coordinate. Along either enlargement the point is the starting point or a disc point with a centre in KK: for a nested-disc point an enlargement exceeding its limiting radius contains a member of the defining sequence and is centred there. Thus neither path passes through xx. These continuous enlargement paths are the disc homotopy of Proposition 4.21.2 in a disc coordinate, or equivalently the Gauss enlargement of Proposition 4.3.1. The punctured neighbourhoods are nonempty and give one branch.

It remains to check the model assertion when the supplied model X\mathfrak X is an arbitrary admissible semistable formal model of XanX^{\mathrm{an}}. Return to a component vertex xx on this model and write η=sp⁡X(x)\eta=\operatorname{sp}_{\mathfrak X}(x) for the generic point of its component, C={η}‾redC=\overline{\{\eta\}}_{\rm red}, and F=H(x)~F=\widetilde{\mathcal H(x)}. We use only that the special fibre is a reduced nodal curve and that the generic fibre is the given analytic curve. The following collars are taken on the split algebraic model realizing xx, which we call the reference model.

On an affine formal chart containing η\eta, the prime of the reduction defined by xx is the minimal prime of CC. Proposition 4.12.2 gives the unique point over η\eta; Proposition 4.24.2 identifies the fraction field of the reduced chart component with FF by evaluation. Restriction to a smaller chart gives the same map on the function field, so k(C)=Fk(C)=F canonically.

We verify completeness of the normalized component without a properness assumption on an individual affine chart. Apply Theorem 3.19.25 at xx with the whole quasi-compact quasi-separated strictly analytic curve as neighbourhood, first with X\mathfrak X and then with the completed reference model. It identifies the two spaces of valuations with centre on the respective reduced components, over the space of valuations of F/kF/k. Here the reduced closure in each application is exactly the indicated component. The reference component is proper, so every valuation of F/kF/k has exactly one centre on it, by the valuative criterion for properness [StacksAlg] (Tag 0BX5, (1) to (3)). Consequently every such valuation has exactly one centre on CC. Its normalization N→CN\to C is finite and birational [StacksAlg] (Tag 0BXR, (1) and (3)), hence proper [StacksAlg] (Tag 02OG, (1) to (2)), since CC is locally noetherian. Lemma 3.19.23 applied to N→CN\to C thus gives exactly one centre on NN as well.

For clarity, this centre comparison identifies NN with the smooth projective curve CxC_x, over FF. At a closed point of either normal curve the local ring is a DVR [StacksAlg] (Tag 00PD, (5) to (1)). A nontrivial valuation ring of F/kF/k dominating this DVR equals it: a uniformizer has positive value and every element of F×F^\times is a unit times a power of that uniformizer. Each such valuation has a unique centre on each curve, and the centre cannot be generic. Thus the identification by valuations matches the closed points and their local rings inside FF. Both curves are integral noetherian curves of finite type over kk; their proper closed subsets are finite sets of closed points. This gives a homeomorphism, including the generic point. On any nonempty open of an integral scheme, the regular functions are the intersection of its local rings in its function field (the local representatives agree on overlaps). The homeomorphism therefore gives the unique isomorphism inducing the identity on FF. Denote the normalization map by ν:Cx→C\nu:C_x\to C.

We now show that the branch indexed by p∈Cx(k)p\in C_x(k) actually specializes to ν(p)\nu(p) near xx on X\mathfrak X. Choose a finite formal affine cover of X\mathfrak X, as in the proof of Theorem 3.19.25. The affinoid generic fibres UU are compact in the Hausdorff curve. Near xx we can discard the finitely many which miss xx. For each remaining chart Spf⁡A0\operatorname{Spf} A_0, lift finitely many kk-algebra generators of A0/K∘∘A0A_0/K^{\circ\circ}A_0 to ti∈A0t_i\in A_0. Write ψi∈F\psi_i\in F for their images on the reduced component. Choose hi∈K(X)h_i\in K(X) representing ψi\psi_i at xx, taking hi=0h_i=0 if ψi=0\psi_i=0. For this choice, write each nonzero ψi\psi_i on an affine open of the reference component as a quotient of residues of two functions on a reference algebraic model affine, with nonzero denominator at the component generic point. Lifting that quotient gives hih_i of norm one at xx. On a dense open of the reference component the numerator and denominator are regular and the denominator has nonzero reduction. There hih_i is a regular model function after inverting the denominator. The functions hih_i are analytic on an ordinary neighbourhood of xx: its algebraic support is generic since a closed point of XX is KK-rational and cannot have type II. Continuity on UU now gives ∣ti−hi∣<1on U near x.|t_i-h_i|<1 \qquad\hbox{on \(U\) near \(x\)}. A single ordinary neighbourhood works on the corresponding chart intersections for all members of the finite list.

For a nonzero rational germ hh of norm one at xx, the reference branch collar with coordinate zz, ∣z∣→1|z|\to1, satisfies ∣h(y)∣=∣z(y)∣ord⁡p(h~x)on a sufficiently thin full collar.|h(y)|=|z(y)|^{\operatorname{ord}_p(\widetilde h_x)} \quad\hbox{on a sufficiently thin full collar.} Indeed hh is a unit on a neighbourhood of xx; cofinality of the reference stars puts the whole thin collar there. Theorem 4.24.5 gives the formula on the Gauss segment and Corollary 4.10.3 extends it to all points of the collar.

Fix pp and a chart UU. If ν(p)\nu(p) lies outside its special affine, some nonzero ψi\psi_i has a pole at pp: otherwise the coordinate ring of the affine piece of CC, generated by the ψi\psi_i, lies in OCx,p\mathcal O_{C_x,p}. This would give the same valuation a centre in the affine piece, contrary to uniqueness. For this ii, ∣hi∣>1|h_i|>1 on the thin collar, which excludes UU there by ∣ti∣≤1|t_i|\le1 and ∣ti−hi∣<1|t_i-h_i|<1. If q=ν(p)q=\nu(p) lies in the special affine, lift the values ψi(p)∈k\psi_i(p)\in k to ci∈K∘c_i\in K^\circ. When ψi−ψi(p)\psi_i-\psi_i(p) vanishes in FF, ∣hi−ci∣<1|h_i-c_i|<1 near xx by continuity. Otherwise its order at pp is positive, so the collar formula gives ∣hi−ci∣<1|h_i-c_i|<1 on the thin collar. Consequently ∣ti−ci∣<1|t_i-c_i|<1 there on UU for every ii. These inequalities cut out the tube of the single closed point qq in this chart, since the reduced coordinate values of the generators determine the point. Finiteness of the cover gives the claim on a whole collar of the branch.

Finally the cofinal stars can respect these specializations simultaneously. Remove from CxC_x the finitely many reference node branches, the complements of the preimages of the special affines just used, and the complements of the dense smooth reference opens on which all hih_i are regular model functions with the specified reductions. These exceptions are finite since each of the chosen opens contains the component generic point. For a remaining pp, throughout its reference tube the reduction of hih_i is its value ψi(p)\psi_i(p): in a model affine hi−cih_i-c_i reduces to zero at the specialization. Hence within the common ordinary neighbourhood from above, every point of that tube reduces on X\mathfrak X to ν(p)\nu(p), in whichever chart contains it. Take a reference star contained in this neighbourhood and in any prescribed neighbourhood of xx, and thin its finitely many exceptional branches by the preceding collar argument. Theorem 4.25.3 supplies such stars with connected punctured pieces. Each piece is now in the formal tube of its specified point. Lemma 4.24.1 on the reduced nodal special fibre identifies the points above a node with its two geometric branches (the residue field is algebraically closed). At a self-node these are distinct points of this same CxC_x; their punctured-star pieces are disjoint even though both specialize to that node. ◻

For the Gauss point on P1\mathbb P^1, Cx=PK~1C_x=\mathbb P^1_{\widetilde K}: the finite residual points give the inward discs and the point at infinity gives the outward branch. On the closed unit disc, that outward branch is absent from the domain. On a general curve the residual curve can have positive genus. Replacing it by P1\mathbb P^1 would discard precisely the geometry that residual degree and harmonicity can detect.

4.6 Semistable vertex sets and formal models

Definition 4.6.1. A semistable vertex set of XanX^{\mathrm{an}} is a finite set VV of type II points such that Xan∖VX^{\mathrm{an}}\setminus V is a disjoint union of open balls and finitely many open annuli. Its skeleton Σ(X,V)\Sigma(X,V) is VV together with the skeleta of those annuli. The length of an annular edge is its logarithmic modulus.

Theorem 4.6.2. A smooth proper connected curve over the specified algebraically closed complete field has semistable vertex sets. Semistable formal models correspond to such vertex sets. A model dominates another exactly when its vertex set contains the other one. The associated finite metric graph is a skeleton of the analytic curve, and the local retractions glue to a deformation retraction onto it.

Proof. Let O=K∘O=K^\circ, k=K~k=\widetilde K, and choose π∈O\pi\in O with 0<∣π∣<10<|\pi|<1. Write CC for the given smooth proper connected curve and X=CanX=C^{\mathrm{an}}. The curve is projective [StacksAlg] (Tag 0A26). Fix any admissible semistable formal model X\mathfrak X with its identification Xη=X\mathfrak X_\eta=X. We will apply the correspondence for algebraic completions by proving that this very model is such a completion over XX. Its special fibre D=XsD=\mathfrak X_s is a reduced curve with at worst nodes. No pointed chart on X\mathfrak X is a hypothesis of the argument.

The formal model has a finite cover by formal affines: apply the finite chart argument in Theorem 3.19.25 to the whole compact strictly analytic curve. Their generic fibres form an analytic covering with the neighbourhood property, and specialization is surjective. Hence DD is of finite type over kk. For every irreducible component its generic point has a unique analytic lift by Proposition 4.12.2; this is a type II point by Proposition 4.24.2. The formal-model clause of Theorem 4.5.1 identifies the normalization of that component with its smooth projective residual curve. Thus the disjoint union PP of the normalizations is proper over kk, and P→DP\to D is finite surjective by [StacksAlg] (Tag 0BXR) (or Lemma 4.24.1). In particular DD is proper. Indeed, after any base change the preimage of a closed subset of DD under this surjection has the same image in the base as the closed subset, so DD is universally closed. The finite morphism P×kP→D×kDP\times_kP\to D\times_kD maps the closed diagonal of PP onto the image of the diagonal of DD. The latter image is closed. The diagonal is a closed immersion locally around its image, on U×kUU\times_k U for affine opens U⊂DU\subset D; it is therefore a closed immersion. This proves separatedness as well.

Put Rn=O/πnOR_n=O/\pi^nO and let DnD_n be the nn-th scheme truncation of X\mathfrak X, for n≥1n\geq1. These form a cartesian tower, with Dn×Rnk=DD_n\times_{R_n}k=D. The ideal K∘∘/πnOK^{\circ\circ}/\pi^nO is nil since KK has rank one. Its extension to any ring is nil, so passage to the closed fibre gives a homeomorphism also after any base change on RnR_n, for the base, the source and its fibre square. Universal closedness for DnD_n follows from that for DD after these identifications. The diagonal image of DnD_n is likewise closed. The same affine diagonal argument over RnR_n gives separatedness.

On a formal affine Spf⁡A0\operatorname{Spf} A_0 admissibility gives a surjection O⟨T1,…,Tm⟩→A0O\langle T_1,\ldots,T_m\rangle\to A_0 and π\pi-torsion-freeness. Every nonzero element of OO divides some πj\pi^j, so A0A_0 is OO-torsion-free, hence flat [StacksAlg] (Tag 0539). Thus B=A0/πnA0B=A_0/\pi^n A_0 is flat over RnR_n. It is a quotient of S=Rn[T1,…,Tm]S=R_n[T_1,\ldots,T_m], because a restricted series reduces to a polynomial. To check finite presentation, apply [StacksAlg] (Tag 05IK) on Spec⁡S→Spec⁡Rn\operatorname{Spec}S\to\operatorname{Spec}R_n to the cyclic SS-module BB. The polynomial morphism is finitely presented, the module sheaf is of finite type, and the base has only one point, so its weakly associated points are locally finite. Every stalk of this sheaf is RnR_n-flat: localization is a filtered colimit of copies of the flat module BB. The theorem gives finite presentation on the entire affine space. Consequently the kernel of S→BS\to B is finitely generated, and BB is a finitely presented RnR_n-algebra. The finite affine cover and separatedness now give properness, flatness and finite presentation of every Dn/RnD_n/R_n.

Proposition B.1.2 and Theorem B.1.3 apply to precisely this tower. They give a projective flat finitely presented OO-scheme YY, at worst nodal of relative dimension one, together with compatible isomorphisms at all levels. On formal affines the rings and their restriction sheaves are the π\pi-adic inverse limits of those truncations. We obtain an isomorphism Y^π≃X\widehat Y_\pi\simeq\mathfrak X and, by Theorem 4.13.2, the specified analytic isomorphism (YK)an≃Can(Y_K)^{\mathrm{an}}\simeq C^{\mathrm{an}}.

This last identification comes from an isomorphism YK≃CY_K\simeq C. Here is the precise use of Theorem 2.34.8. For projective KK-schemes L,ML,M, embed MM in projective space. An analytic map Lan→ManL^{\mathrm{an}}\to M^{\mathrm{an}} pulls back the hyperplane line bundle and its generating sections. The theorem algebraizes the line bundle and the sections; the evaluation map is surjective because its coherent cokernel has zero analytification. They define a map to projective space [StacksAlg] (Tag 01NE). Its analytification is the original projective-space map, as one sees on each coordinate chart by taking ratios of the sections. Each homogeneous equation of MM pulls back to an algebraic section of a tensor power whose analytification vanishes, hence to zero. The map therefore factors through MM scheme-theoretically. Two such algebraic maps with the same analytification have isomorphic bundles with the same ordered sections by full faithfulness, so are equal. Applying this argument to the isomorphism and its inverse gives the assertion, with the prescribed identification.

In particular YKY_K is smooth. Theorem 4.18.2 gives split pointed charts at the finitely many special nodes: its finite extension is KK, and smoothness of YKY_K makes every smoothing parameter nonzero by Theorem 4.15.3. Smooth closed special points have residue field kk, which is algebraically closed. Thus YY, with its generic identification, is a split nodal algebraic model of CC. The formal isomorphism over XX identifies the specialization maps and the component vertices.

Theorem 4.38.2, with no markings, supplies an initial model. Theorem 4.42.5 now applies to every model under consideration: it realizes every semistable vertex set, identifies the arrows inducing the identity on XX with reverse inclusions, and identifies the graph with its original logarithmic edge lengths. The local disc and annulus retractions glue by Theorem 4.23.3; transport along the formal isomorphism gives the asserted deformation retraction. ◻

The special fibre explains the gluing. Tubes of smooth closed points are open balls; tubes of ordinary double points are open annuli; generic points of components lift to the vertices. The annular modulus comes from an equation UV=aUV=a, and the incidence of components gives the incidence of edges. This reconstructs a metric graph from a model, while Theorem 4.6.2 supplies the converse construction.

Adding a type II point in an annular edge subdivides the edge without changing its total length. Adding a vertex in a ball attaches a new tree segment. The skeleton therefore depends on the chosen model, although the metric on common paths is compatible. For punctured curves, one chooses a vertex set compatible with the finitely many punctures and adds the corresponding infinite rays. Over a non-algebraically-closed field, semistable reduction can require a finite extension; descent of models, weights and integrals must then be handled separately.

4.7 The slope formula and its sign

Let f∈K(X)×f\in K(X)^\times and put F=−log⁡∣f∣F=-\log|f| away from its zeros and poles. At a type II point xx, choose c∈K×c\in K^\times with ∣cf(x)∣=1|cf(x)|=1. The residue of cfcf is a nonzero rational function f~x\widetilde f_x on CxC_x, defined up to a nonzero residual constant. Its divisor is therefore unambiguous.

Theorem 4.7.1. (Slope formula). For a direction vv at a type II point, corresponding to pv∈Cx(K~)p_v\in C_x(\widetilde K), dvF(x)=ord⁡pv(f~x).d_vF(x)=\operatorname{ord}_{p_v}(\widetilde f_x). Only finitely many slopes are nonzero, and their sum is zero. On a skeleton of the punctured curve containing all zeros and poles among its ends, FF factors through the retraction and is piecewise affine with integral slopes. Toward a zero or pole aa, with ray parameter t=−log⁡∣z∣t=-\log|z| for a local parameter zz, it has slope ord⁡a(f)\operatorname{ord}_a(f).

Open proof obligation (RC2). Prove equality of residual orders and analytic slopes in the stated scope; BPRStructure Theorem 5.15 supplies it.

NAG-900007 · Proof status

Example 4.7.2. Take f=Tf=T at the unit Gauss point of P1\mathbb P^1. Its residual function is tt, with divisor [0]−[∞][0]-[\infty]. The outgoing slopes of −log⁡∣T∣-\log|T| are +1+1 toward the residual zero, −1-1 toward infinity, and zero in all other directions. For log⁡∣T∣\log|T|, both signs reverse. The later identity d′d′′[log⁡∣f∣]=[div⁡(f)]d'd''[\log|f|]=[\operatorname{div}(f)] uses the latter logarithm; graph Laplacian conventions must be matched to it, rather than inferred from the word “slope”.

4.8 Higher-dimensional skeleta and tropical coordinates

The curve construction has a higher-dimensional local model, but existence of such a model is an additional hypothesis. Suppose a formal model is strictly semistable, with local ètale charts over Spf⁡K∘⟨X0,…,Xr,T1,…,Td−r⟩/(X0⋯Xr−a),0<∣a∣<1.\operatorname{Spf}K^\circ\langle X_0,\ldots,X_r,T_1,\ldots,T_{d-r}\rangle /(X_0\cdots X_r-a),\qquad 0<|a|<1. Its standard skeleton has coordinates ui=−log⁡∣Xi∣≥0u_i=-\log|X_i|\geq 0 with ∑iui=v(a)\sum_i u_i=v(a). The Gauss-point construction and residue calculation were given in Section 3.8. The simplex has real dimension rr, while the generic fibre has analytic dimension dd.

Definition 4.8.1. (Formal strictly semistable pair). In the setting of [GRW16] (Definition 3.7), take a connected quasi-compact admissible formal scheme with the local charts above, together with distinguished effective Cartier divisors HiH_i. Each HiH_i has irreducible support on the generic fibre and, in each chart, is either absent or defined by one of the free coordinates TjT_j. Their sum HH gives a formal strictly semistable pair (X,H)(\mathfrak X,H). The irreducibility condition excludes self-intersections of an individual horizontal component.

When ss horizontal coordinates vanish on the stratum under consideration, the corresponding canonical polyhedron is Δ(r,a)×R≥0s,Δ(r,a)={(u0,…,ur)≥0:∑iui=v(a)}.\Delta(r,a)\times\mathbb R_{\geq0}^{s},\qquad \Delta(r,a)=\{(u_0,\ldots,u_r)\geq0:\textstyle\sum_i u_i=v(a)\}. The unbounded coordinates are −log⁡∣T1∣,…,−log⁡∣Ts∣-\log|T_1|,\ldots,-\log|T_s| on the complement of the horizontal divisor. The remaining free coordinates have Gauss radius one. Thus the polyhedron has dimension r+sr+s, while the associated residual stratum has dimension d−r−sd-r-s. Both enter a higher-dimensional slope formula.

Strata index the canonical polyhedra, and their incidence determines the gluing. Writing X=XηX=\mathfrak X_\eta, GRW constructs the skeleton in X∖∣Hη∣X\setminus|H_\eta| and its canonical retraction; see [GRW16] (§§3–5, especially Theorem 4.13). This is an external geometric theorem, extending the explicit annulus construction. The availability of the local equation does not establish a global strictly semistable model for every variety. Alterations and extensions, when used to obtain models, require their own descent arguments.

Definition 4.8.2. On an analytic domain VV, invertible analytic functions f1,…,fmf_1,\ldots,f_m define an analytic moment map f:V→(Gmm)anf:V\to(\mathbb G_m^m)^{\mathrm{an}} and its tropicalization ftrop⁡(x)=(−log⁡∣f1(x)∣,…,−log⁡∣fm(x)∣).f_{\operatorname{trop}}(x)=(-\log|f_1(x)|,\ldots,-\log|f_m(x)|). This map need not be injective or have image of the same dimension as VV. Its coordinates present real functions; the analytic geometry supplies the weights and the admissible local dimensions.

We now pass from this analytic foundation to tropical geometry. Three constructions must be kept distinct: a skeleton attached to a model, the tropical image of a chosen moment map, and the tropical skeleton selected by Shilov points of tropical fibres. GRW relates them under precise hypotheses; the relations, degree formulas and faithful embeddings are developed in Chapter 6. Their differential applications start only after the lattice multiplicities and refinement rules of Chapter 5 have been established.

4.9 Finite Galois descent of the analytic topology

Fix a complete non-Archimedean field KK with its given absolute value, and a finite Galois extension L/KL/K with group GG. Give LL the unique extending absolute value of Lemma 1.12.1. We use the scalar extension constructed in Theorem 2.40.12. The action of GG on XLX_L is semilinear over LL and fixes its projection p:XL→Xp:X_L\to X.

Lemma 4.9.1. NAG-002023 (Galois action on completed fibres). For a KK-analytic space XX and x∈Xx\in X, the fibre p−1(x)p^{-1}(x) is a nonempty finite GG-orbit. More precisely, putting H=H(x)H=\mathcal H(x), there is a finite product decomposition H⊗^KL=H⊗KL=∏j=1sEj,∑j=1s[Ej:H]=[L:K],H\widehat\otimes_K L=H\otimes_K L=\prod_{j=1}^s E_j, \qquad \sum_{j=1}^s[E_j:H]=[L:K], where the EjE_j are finite field extensions of HH, equipped with their unique extending absolute values. The factors correspond to the points yjy_j of the fibre, with H(yj)≃Ej\mathcal H(y_j)\simeq E_j. The GG-action permutes the factors transitively and preserves their valued-field structures.

Proof. Choose a KK-basis of LL. Its coordinate maps are bounded by Lemma 1.12.1. Tensoring these finitely many maps with HH compares the projective tensor norm with a coordinate maximum norm on H[L:K]H^{[L:K]} in both directions. The latter is complete. Thus the algebraic tensor is already complete and is the displayed completed tensor.

For a KK-affinoid algebra AA, the same coordinate comparison identifies A⊗^KLA\widehat\otimes_K L with the finite free AA-algebra A⊗KLA\otimes_K L in its canonical finite-module topology. It is KK-affinoid by Theorem 2.33.18. Every bounded KK-point restricts on LL to its unique extending absolute value, so its KK-spectrum and LL-spectrum agree. The affinoid fibre formula therefore applies to the projection and gives precisely H⊗KLH\otimes_K L. These identifications glue on the chart atlas.

By the primitive-element lemma, [StacksAlg] (Tag 030N), write L=K[α]L=K[\alpha] with separable minimal polynomial PP. The coprimality of PP and P′P' persists in H[T]H[T] by the same polynomial Bezout identity. Factoring PP into distinct monic irreducibles over HH and applying the Chinese remainder construction gives the product of fields Ej=H[T]/(Pj)E_j=H[T]/(P_j). The sum of their degrees is deg⁡P=[L:K]\deg P=[L:K]. Each field is complete and has precisely one extending absolute value by Lemma 1.12.1. The fibre construction of Theorem 2.40.13 now gives the point and completed-residue-field assertions.

The fixed algebra is (H⊗KL)G=H(H\otimes_K L)^G=H. Indeed, write an invariant element as ∑ihi⊗ℓi\sum_i h_i\otimes\ell_i, with the finitely many hih_i linearly independent over KK. Applying each σ−1\sigma-1 shows that every ℓi\ell_i lies in LG=KL^G=K, where the last equality is [StacksAlg] (Tag 09DW). If the primitive idempotents of the product had two or more GG-orbits, the sum of the idempotents in one orbit would be a nontrivial idempotent in this fixed field. This is impossible. Hence the action is transitive. An induced HH-isomorphism between two factors preserves their absolute values by uniqueness, which completes the proof. ◻

Theorem 4.9.2. NAG-002024 (The scalar-extension quotient). For every KK-analytic space XX, the scalar-extension projection is a continuous, open, closed surjection and induces a homeomorphism XL/G → ∼  X.X_L/G\ \xrightarrow{\ \sim\ }\ X. For every GG-invariant subset S⊆XLS\subseteq X_L, its induced map S/G→p(S)S/G\to p(S) is also a homeomorphism, with the subspace topologies. For every topological space TT, the map p×idTp\times\mathrm{id}_T is an open quotient map.

Proof. On an affinoid chart V=M(A)V=\mathcal M(A), its inverse image is VL=ML(A⊗^KL)V_L=\mathcal M_L(A\widehat\otimes_K L). The field automorphisms of LL are isometries, so their coefficient actions extend to bounded algebra automorphisms and hence continuous maps of VLV_L. These actions agree on overlaps and give a continuous GG-action on XLX_L. By Lemma 4.9.1, the projection is surjective and its fibres are exactly the orbits.

We verify closedness using the affinoid quasi-net atlas. Let CC be closed in XLX_L. For each affinoid chart VV, the set C∩VLC\cap V_L is closed in the compact space VLV_L; its image p(C)∩Vp(C)\cap V is compact and thus closed in the Hausdorff space VV. The quasi-net condition now makes p(C)p(C) closed in XX. To see this directly, for x∉p(C)x\notin p(C) choose finitely many chart members ViV_i containing xx whose union contains a neighborhood NN of xx. Each relative open complement Vi∖p(C)V_i\setminus p(C) contains xx, so it contains Oi∩ViO_i\cap V_i for an ambient open neighborhood OiO_i of xx. The neighborhood N∩⋂iOiN\cap\bigcap_i O_i is disjoint from p(C)p(C). This argument does not require XX itself to be Hausdorff or good.

A closed continuous surjection is a quotient map. Since its fibres are the GG-orbits, it gives the stated homeomorphism. It is open as well: for any open U⊂XLU\subset X_L, the set p−1(p(U))=⋃g∈GgUp^{-1}(p(U))=\bigcup_{g\in G}gU is open, so p(U)p(U) is open by the quotient property.

A GG-invariant subset is saturated for pp. If F⊂SF\subset S is relatively closed, write F=C∩SF=C\cap S with CC closed in XLX_L. Saturation gives p(F)=p(C)∩p(S)p(F)=p(C)\cap p(S), which is relatively closed. Thus S→p(S)S\to p(S) is a quotient map with orbit fibres and yields the second homeomorphism. Finally, p×idTp\times\mathrm{id}_T sends a basic open product U×WU\times W to the open product p(U)×Wp(U)\times W. It is a continuous open surjection and therefore a quotient map. ◻

Corollary 4.9.3. NAG-002025 (Descent of equivariant retractions). Let S⊆XLS\subseteq X_L be GG-invariant and let HL:XL×[0,1]→XLH_L:X_L\times[0,1]\to X_L be a continuous GG-equivariant strong deformation retraction onto SS. There is a unique continuous strong deformation retraction H:X×[0,1]→XH:X\times[0,1]\to X onto p(S)p(S) such that H(p(y),t)=p(HL(y,t)).H(p(y),t)=p(H_L(y,t)). Writing rL=HL(−,1)r_L=H_L(-,1) and r=H(−,1)r=H(-,1), a continuous function F:X→RF:X\to\mathbf R satisfies F∘p=(F∘p)∘rLF\circ p=(F\circ p)\circ r_L if and only if F=F∘rF=F\circ r.

Proof. Equivariance makes p∘HLp\circ H_L constant on every fibre of p×id[0,1]p\times\mathrm{id}_{[0,1]}. The quotient property in Theorem 4.9.2 produces a unique continuous HH. Its value at t=0t=0 is the identity, its value at t=1t=1 lies in p(S)p(S), and it fixes every point of p(S)p(S) throughout the homotopy: choose a lift in SS and use the same assertions for HLH_L. Finally p∘rL=r∘pp\circ r_L=r\circ p and surjectivity of pp prove the equivalence for FF. ◻

For an analytic unit ff on a KK-analytic domain UU, apply the corollary with X=UX=U and an equivariant strong deformation retraction of ULU_L. Its scalar extension satisfies ∣fL(y)∣=∣f(p(y))∣|f_L(y)|=|f(p(y))| by evaluation, so the last assertion applies to F=−log⁡∣f∣F=-\log|f| on UU. All absolute values here are the fixed original absolute value and its extensions; the logarithmic values are unchanged by descent.

4.10 Laurent coefficients and local slopes

Let KK be a complete non-Archimedean field, with arbitrary characteristic and possibly trivial absolute value. For 0<r≤R0<r\leq R, put Ar,R=K{R−1T,rT−1},f=∑n∈ZanTn,∥f∥r,R=max⁡n∈Z∣an∣max⁡(rn,Rn).\begin{gathered} A_{r,R}=K\{R^{-1}T,rT^{-1}\}, \qquad f=\sum_{n\in\mathbb Z}a_nT^n,\\ \|f\|_{r,R}=\max_{n\in\mathbb Z}|a_n|\max(r^n,R^n). \end{gathered} The coefficient description and its multiplication are supplied by Theorem 2.13.2 and Proposition 2.13.3. For r≤ρ≤Rr\leq\rho\leq R, write ηρ=ζ0,ρ\eta_\rho=\zeta_{0,\rho} and Mf(ρ)=∣f(ηρ)∣=max⁡n∣an∣ρn,Ff(t)=−log⁡Mf(e−t).M_f(\rho)=|f(\eta_\rho)|=\max_n|a_n|\rho^n, \qquad F_f(t)=-\log M_f(e^{-t}). For f≠0f\ne 0, the maximum is positive and attained, and the set of maximizing exponents is finite. Denote its least and greatest elements by mf(ρ)m_f(\rho) and Mfexp(ρ)M_f^{\mathrm{exp}}(\rho), respectively.

Lemma 4.10.1. NAG-002026 (Extreme Laurent exponents). For f,g∈Ar,R∖{0}f,g\in A_{r,R}\setminus\{0\} and r≤ρ≤Rr\leq\rho\leq R, mfg(ρ)=mf(ρ)+mg(ρ),Mfgexp(ρ)=Mfexp(ρ)+Mgexp(ρ).m_{fg}(\rho)=m_f(\rho)+m_g(\rho),\qquad M_{fg}^{\mathrm{exp}}(\rho) =M_f^{\mathrm{exp}}(\rho)+M_g^{\mathrm{exp}}(\rho). If r<ρ<Rr<\rho<R and t0=−log⁡ρt_0=-\log\rho, the one-sided derivatives satisfy Ff,+′(t0)=mf(ρ),Ff,−′(t0)=Mfexp(ρ).F'_{f,+}(t_0)=m_f(\rho),\qquad F'_{f,-}(t_0)=M_f^{\mathrm{exp}}(\rho).

Proof. Write f=∑aiTif=\sum a_iT^i and g=∑bjTjg=\sum b_jT^j, and fix ρ\rho. The weighted coefficient families tend to zero in both directions. In the coefficient of Tmf+mgT^{m_f+m_g}, the product amfbmga_{m_f}b_{m_g} is the unique summand of weighted size Mf(ρ)Mg(ρ)M_f(\rho)M_g(\rho). Indeed, a summand of that size must use a maximizing exponent of each factor, and neither exponent can be smaller than the corresponding minimum. All other summands have strictly smaller size. Their sizes tend to zero, so their supremum is strictly smaller as well. The ultrametric inequality therefore gives a coefficient of that weighted size.

For a smaller exponent, every convolution summand has strictly smaller weighted size, and the same finite-tail argument applies. Gauss multiplicativity, Proposition 1.2.2, now gives the first formula. Choosing the greatest maximizing exponents gives the second formula.

By Proposition 4.4.1, FfF_f is the minimum of finitely many affine functions v(an)+ntv(a_n)+nt on the compact parameter interval. Near t0t_0, only those attaining the minimum at t0t_0 can contribute. To the right their smallest slope is selected; to the left their greatest slope is selected. This proves both derivative formulas. ◻

Proposition 4.10.2. NAG-002027 (Annular units). An element f∈Ar,Rf\in A_{r,R} is a unit if and only if f=cTm(1+h),c∈K×,m∈Z,h∈Ar,R,∥h∥r,R<1.f=cT^m(1+h),\qquad c\in K^\times,\quad m\in\mathbb Z,\quad h\in A_{r,R},\quad \|h\|_{r,R}<1. The exponent mm is unique, and cc may be taken to be the coefficient ama_m of ff. At every analytic point xx of the annulus, ∣f(x)∣=∣c∣ ∣T(x)∣m.|f(x)|=|c|\,|T(x)|^m.

Proof. Suppose fg=1fg=1. Lemma 4.10.1 gives, at every radius, (Mfexp−mf)+(Mgexp−mg)=0.\bigl(M_f^{\mathrm{exp}}-m_f\bigr) +\bigl(M_g^{\mathrm{exp}}-m_g\bigr)=0. Both terms are nonnegative integers, so ff has a unique maximizing exponent at every radius. Only finitely many exponents can maximize anywhere on [r,R][r,R], as in the proof of Proposition 4.4.1. A unique maximizer is therefore locally constant in the radius. Connectedness of [r,R][r,R] gives one exponent mm throughout; the same conclusion is immediate if r=Rr=R.

Set c=amc=a_m and h=f/(cTm)−1h=f/(cT^m)-1. At either endpoint every coefficient term of hh has size strictly less than one. The terms tend to zero, so their maximum is strictly less than one. The coefficient norm is the maximum of the two endpoint Gauss norms; hence ∥h∥r,R<1\|h\|_{r,R}<1. Conversely, Lemma 1.1.4 inverts 1+h1+h, so the displayed form is a unit.

For any analytic point, Lemma 1.6.1 gives ∣h(x)∣<1|h(x)|<1, whence ∣1+h(x)∣=1|1+h(x)|=1. This proves the norm identity. Any displayed form has mm as its unique maximizing exponent at every radius, which proves uniqueness. ◻

Corollary 4.10.3. NAG-002028 (Units on standard open domains). An analytic unit on a standard open annulus r<∣T∣<Rr<|T|<R, where 0≤r<R≤∞0\leq r<R\leq\infty, has ∣f(x)∣=∣c∣ ∣T(x)∣m(c∈K×, m∈Z).|f(x)|=|c|\,|T(x)|^m\qquad(c\in K^\times,\ m\in\mathbb Z). Thus its absolute value factors through the radial retraction. An analytic unit on a standard open or closed disc of positive radius with centre a∈Ka\in K has constant absolute value ∣f(a)∣|f(a)|.

Proof. Restrict an annular unit and its inverse to closed subannuli of positive modulus. Proposition 4.10.2 applies to each restriction. When two radius intervals overlap in a nonempty open interval, the affine functions log⁡∣c∣+mlog⁡ρ\log|c|+m\log\rho agree there, so their slope and constant agree. Any two compact radius intervals lie in a common larger compact subinterval of (r,R)(r,R); comparing with that subinterval gives the asserted global formula.

For a closed disc of radius q>0q>0, translate the centre to zero and write f=∑n≥0anTnf=\sum_{n\geq 0}a_nT^n and its inverse g=∑n≥0bnTng=\sum_{n\geq 0}b_nT^n. Restriction to the radius-qq circle identifies these coefficient series with elements of Aq,qA_{q,q}. Their greatest maximizing exponents are nonnegative and sum to zero by Lemma 4.10.1. Both are zero. Consequently a0≠0a_0\ne 0 and ∥f/a0−1∥q<1\|f/a_0-1\|_q<1, which gives ∣f(x)∣=∣a0∣|f(x)|=|a_0| throughout the disc. Restricting an open-disc unit to each closed subdisc centred at aa proves the open-disc assertion. ◻

Proposition 4.10.4. NAG-002029 (Residual orders along Gauss-circle paths). Let f=∑anTn∈Ar,R∖{0}f=\sum a_nT^n\in A_{r,R}\setminus\{0\}, let r<ρ<Rr<\rho<R with ρ∈∣K×∣\rho\in|K^\times|, and choose b,c∈K×b,c\in K^\times with ∣b∣=ρ|b|=\rho and ∣c∣=Mf(ρ)|c|=M_f(\rho). Put k=K~k=\widetilde K and dn=c−1anbn,Pf(Z)=∑ndn~Zn∈k[Z,Z−1].d_n=c^{-1}a_nb^n, \qquad P_f(Z)=\sum_n\widetilde{d_n}Z^n\in k[Z,Z^{-1}]. Then PfP_f is a nonzero Laurent polynomial representing the residue of c−1fc^{-1}f in H(ηρ)~=k(T/b~)\widetilde{\mathcal H(\eta_\rho)}=k(\widetilde{T/b}). For sufficiently small s≥0s\geq 0, consider the paths γ0(s)=ζ0,ρe−s,γ∞(s)=ζ0,ρes,γα(s)=ζbu,ρe−s(α∈k×, u∈K∘, u~=α).\begin{gathered} \gamma_0(s)=\zeta_{0,\rho e^{-s}},\qquad \gamma_\infty(s)=\zeta_{0,\rho e^s},\\ \gamma_\alpha(s)=\zeta_{bu,\rho e^{-s}} \quad\left(\alpha\in k^\times,\ u\in K^\circ,\ \widetilde u=\alpha\right). \end{gathered} They start at ηρ\eta_\rho, and dds∣0+(−log⁡∣f(γα(s))∣)=ord⁡α(Pf)(α∈k∪{∞}).\left.\frac{d}{ds}\right|_{0+} \bigl(-\log|f(\gamma_\alpha(s))|\bigr) =\operatorname{ord}_\alpha(P_f) \qquad\left(\alpha\in k\cup\{\infty\}\right). For α∈k×\alpha\in k^\times and d=ord⁡α(Pf)d=\operatorname{ord}_\alpha(P_f), the stronger identity ∣f(γα(s))∣=∣c∣e−ds|f(\gamma_\alpha(s))|=|c|e^{-ds} holds for all sufficiently small s≥0s\geq 0. The path germ is independent of the lift uu. These statements also apply to a quotient of two nonzero elements of Ar,RA_{r,R} by subtracting their formulas. If kk is algebraically closed, the sum of the displayed outgoing slopes is zero.

Proof. The coefficients dnd_n have size at most one, tend to zero in both directions, and at least one has size one. Hence only finitely many have nonzero reduction, and Pf≠0P_f\ne 0. Such a cc exists because a maximizing term has size ∣anbn∣|a_nb^n|. Laurent polynomials are dense in Ar,RA_{r,R} for a norm dominating ∣⋅∣ηρ|\cdot|_{\eta_\rho}, so H(ηρ)\mathcal H(\eta_\rho) is the completion of the Gauss-valued field K(T)K(T). In the coordinate S=T/bS=T/b, Proposition 3.8.1 therefore identifies its residue field with k(S~)k(\widetilde S). Laurent truncation leaves a tail of norm less than one, proving that the residue of c−1fc^{-1}f is exactly PfP_f.

Its orders at zero and infinity are mf(ρ)m_f(\rho) and −Mfexp(ρ)-M_f^{\mathrm{exp}}(\rho). The corresponding path formulas follow from the two one-sided derivatives in Lemma 4.10.1; the outward path changes t=−log⁡∣T∣t=-\log|T| to t−st-s.

Fix α∈k×\alpha\in k^\times and a lift uu, so ∣u∣=1|u|=1. For every 0<q<10<q<1, substitution of S=u+ZS=u+Z on the closed disc ∣Z∣≤q|Z|\leq q gives c−1f(b(u+Z))=∑j≥0βjZj,βj=∑n∈Zdn(nj)un−j.c^{-1}f\bigl(b(u+Z)\bigr)=\sum_{j\geq0}\beta_jZ^j, \qquad \beta_j=\sum_{n\in\mathbb Z}d_n\binom nj u^{n-j}. Here (nj)\binom nj is the integer binomial coefficient, also for negative nn. Its absolute value in KK is at most one. The inner sum converges by Lemma 1.1.2. Moreover the double family of terms has norm at most ∣dn∣qj|d_n|q^j; for each positive threshold only finitely many pairs exceed it. The same summability lemma therefore permits the interchange of sums. The negative-power expansion follows by the geometric inverse for u+Zu+Z, or equivalently by the integer binomial identity. Thus this is an equality of convergent series, and ∣βj∣≤1|\beta_j|\leq 1.

Reducing each coefficient identifies ∑jβj~Zj\sum_j\widetilde{\beta_j}Z^j with the formal expansion of Pf(α+Z)P_f(\alpha+Z). If d=ord⁡α(Pf)d=\operatorname{ord}_\alpha(P_f), then ∣βd∣=1|\beta_d|=1 and ∣βj∣<1|\beta_j|<1 for j<dj<d. Choose 0<q0<10<q_0<1 so that ∣βj∣<q0d−j(0≤j<d).|\beta_j|<q_0^{d-j}\qquad(0\leq j<d). For q0<q<1q_0<q<1, the ddth term has Gauss size qdq^d. Each lower term has smaller size by this choice, and each higher term has size at most qd+1q^{d+1}. Therefore the Gauss norm is exactly qdq^d. Taking q=e−sq=e^{-s} gives the asserted equality and derivative; at s=0s=0 the equality is ∣f(ηρ)∣=∣c∣|f(\eta_\rho)|=|c|.

If u′u' is another lift of α\alpha, then ∣u−u′∣<1|u-u'|<1. For qq close enough to one, translation by b(u−u′)b(u-u') preserves the radius-ρq\rho q Gauss norm, so the two path germs agree. For a quotient, both numerator and denominator have positive Gauss value and the established formulas hold on a common short initial segment; subtracting logarithms proves the assertion. Finally, writing Pf(Z)=Zmf(ρ)Q(Z),Q(0)≠0,deg⁡Q=Mfexp(ρ)−mf(ρ).P_f(Z)=Z^{m_f(\rho)}Q(Z),\qquad Q(0)\ne0, \qquad\deg Q=M_f^{\mathrm{exp}}(\rho)-m_f(\rho). If kk is algebraically closed, the orders of QQ at its nonzero roots sum to its degree. Adding the orders at zero and infinity gives zero. This proves the slope sum, and subtraction gives the same conclusion for quotients. ◻

All coefficient and Gauss-norm calculations above are unchanged by a complete valued extension of KK that preserves its absolute value. At a radius already in ∣K×∣|K^\times|, the initial Laurent polynomial is carried to the extended residue field coefficientwise, so the same path calculation applies there.

4.11 Implicit functions on entire tubes

The simple-root argument of Lemma 2.54.1 extends to a square system of restricted power series and can be made uniform on increasing affinoid domains. We use its contraction construction and the geometric inversion of Lemma 1.1.4, with the maximum norm for vectors and matrices.

Proposition 4.11.1. NAG-002033 (A uniform implicit-function domain). Let AA be a KK-affinoid algebra over a complete non-Archimedean field, with an admissible submultiplicative Banach norm. Let n≥1n\geq 1, let F=(F1,…,Fn)∈A⟨Z1,…,Zn⟩nF=(F_1,\ldots,F_n)\in A\langle Z_1,\ldots,Z_n\rangle^n have coefficients of norm at most one, and fix z0∈(K∘)nz_0\in(K^\circ)^n. Write J(Z)=(∂Fi/∂Zj)J(Z)=(\partial F_i/\partial Z_j) and put e=F(z0),J0=J(z0),d=det⁡J0,B=A⟨Z1,…,Zn,Q⟩/(F1,…,Fn,Qdet⁡J(Z)−1),p:M(B)⟶M(A).\begin{gathered} e=F(z_0),\qquad J_0=J(z_0),\qquad d=\det J_0,\\ B=A\langle Z_1,\ldots,Z_n,Q\rangle/ (F_1,\ldots,F_n,Q\det J(Z)-1),\\ p:\mathcal M(B)\longrightarrow\mathcal M(A). \end{gathered} Give the following subsets their analytic-domain structures: W={x:∣d(x)∣=1, max⁡i∣ei(x)∣<1},Y={y∈p−1(W):max⁡i∣Zi(y)−z0i∣<1}.\begin{aligned} W&=\{x:|d(x)|=1,\ \max_i|e_i(x)|<1\},\\ Y&=\{y\in p^{-1}(W):\max_i|Z_i(y)-z_{0i}|<1\}. \end{aligned} Then p:Y→Wp:Y\to W is an analytic isomorphism. Its inverse has ZZ-coordinate equal to the unique vector of analytic functions zz on WW satisfying F(z)=0F(z)=0 and max⁡i∣zi(x)−z0i∣<1\max_i|z_i(x)-z_{0i}|<1 at every x∈Wx\in W.

Proof. Translation by z0z_0 is a contractive endomorphism of the unit Tate algebra. Indeed, expand each monomial by the integer binomial formula. Its coefficients have norm at most the norm of the original coefficient. For any positive threshold, only finitely many original coefficients exceed that threshold, and each associated monomial has finitely many terms. The series criterion 1.1.2 therefore permits the sums to be interchanged and gives a restricted series with the same bound. Termwise differentiation is also bounded, since integer coefficients have absolute value at most one. Consequently F(z0+H)=e+J0H+N(H),Ni(H)=∑∣α∣≥2aiαHα,∥aiα∥≤1.F(z_0+H)=e+J_0H+N(H),\qquad N_i(H)=\sum_{|\alpha|\geq2}a_{i\alpha}H^\alpha, \qquad \|a_{i\alpha}\|\leq1. The coefficients of each displayed series tend to zero. In particular all entries of J0J_0, its adjugate, and dd have norm at most one.

The equation ∣d∣=1|d|=1 defines the Laurent affinoid domain D={∣d∣≥1}D=\{|d|\geq 1\}, because ∥d∥≤1\|d\|\leq 1. The set WW is open in DD, and YY is open in p−1(W)p^{-1}(W); domain calculus and base change, Proposition 2.40.5, give their analytic structures. For 0<δ<10<\delta<1, put Wδ={x∈D:max⁡i∣ei(x)∣≤δ},Aδ=A⟨D1,δ−1E1,…,δ−1En⟩/(dD1−1,E1−e1,…,En−en).\begin{aligned} W_\delta&=\{x\in D:\max_i|e_i(x)|\leq\delta\},\\ A_\delta&=A\langle D_1,\delta^{-1}E_1,\ldots, \delta^{-1}E_n\rangle/ (dD_1-1,E_1-e_1,\ldots,E_n-e_n). \end{aligned} In the defining quotient norm, the original coefficients and d−1=D1d^{-1}=D_1 have norm at most one, ∥e∥≤δ\|e\|\leq\delta, and ∥J0−1∥≤1\|J_0^{-1}\|\leq 1 by the adjugate formula. These domains cover WW admissibly: every point lies in the relative interior of some WδW_\delta, by choosing max⁡i∣ei(x)∣<δ<1\max_i|e_i(x)|<\delta<1.

For vectors h,h′h,h' with norms at most δ\delta, telescoping each monomial gives ∥N(h)∥≤δ2,∥N(h)−N(h′)∥≤δ∥h−h′∥.\|N(h)\|\leq\delta^2,\qquad \|N(h)-N(h')\|\leq\delta\|h-h'\|. The same bounds pass to restricted series by convergence. Thus Φ(h)=−J0−1e−J0−1N(h)\Phi(h)=-J_0^{-1}e-J_0^{-1}N(h) preserves the closed radius-δ\delta ball in AδnA_\delta^n and has contraction constant δ<1\delta<1. The iteration from zero used in Lemma 2.54.1 gives a fixed point hδh_\delta with ∥hδ∥≤δ\|h_\delta\|\leq\delta. Set zδ=z0+hδz_\delta=z_0+h_\delta; then F(zδ)=0F(z_\delta)=0 and J(zδ)=J0(I+Vδ),∥Vδ∥≤δ<1.J(z_\delta)=J_0(I+V_\delta),\qquad \|V_\delta\|\leq\delta<1. The matrix geometric inverse gives J(zδ)−1J(z_\delta)^{-1} of norm at most one. Its determinant therefore has norm at most one. Hence Z↦zδZ\mapsto z_\delta and Q↦det⁡J(zδ)−1Q\mapsto\det J(z_\delta)^{-1} define a bounded map B→AδB\to A_\delta and an analytic section sδ:Wδ→Ys_\delta:W_\delta\to Y.

For later use, the monomial telescoping also gives a matrix difference identity F(z0+h)−F(z0+h′)=(J0+R(h,h′))(h−h′),∥R(h,h′)∥≤max⁡(∥h∥,∥h′∥).\begin{gathered} F(z_0+h)-F(z_0+h')=(J_0+R(h,h'))(h-h'),\\ \|R(h,h')\|\leq\max(\|h\|,\|h'\|). \end{gathered} whenever both norms are less than one. Each entry of the remainder contains a monomial of positive degree in h,h′h,h'; the restricted-series tails converge, so this is an identity in the Banach algebra, not only at its analytic points.

If 0<δ1≤δ2<10<\delta_1\leq\delta_2<1, the defining presentations give a contractive restriction Aδ2→Aδ1A_{\delta_2}\to A_{\delta_1}. Apply the difference identity to hδ1h_{\delta_1} and the restricted hδ2h_{\delta_2}. The factor J0+R=J0(I+J0−1R)J_0+R=J_0(I+J_0^{-1}R) is invertible, since ∥J0−1R∥≤δ2<1\|J_0^{-1}R\|\leq\delta_2<1. Hence the roots agree. Analytic morphism gluing gives a section s:W→Ys:W\to Y.

We check the other inverse identity on analytic functions. The domains of YY defined by p(y)∈Wδp(y)\in W_\delta and max⁡i∣Zi(y)−z0i∣≤γ\max_i|Z_i(y)-z_{0i}|\leq\gamma, for 0<δ,γ<10<\delta,\gamma<1, form an admissible affinoid cover. In their defining quotient norms, ∥Z−z0∥≤γ\|Z-z_0\|\leq\gamma and the pulled-back hδh_\delta has norm at most δ\delta. The difference identity becomes 0=F(Z)−F(zδ)=(J0+R)(Z−zδ),∥J0−1R∥≤max⁡(δ,γ)<1.0=F(Z)-F(z_\delta)=(J_0+R)(Z-z_\delta),\qquad \|J_0^{-1}R\|\leq\max(\delta,\gamma)<1. Thus Z=zδZ=z_\delta in the affinoid algebra, and Qdet⁡J(Z)=1Q\det J(Z)=1 identifies the QQ coordinate as well. The base coordinates already agree, so s∘ps\circ p is the identity as an analytic morphism. The argument preserves nilpotents.

Finally, let z′z' be another analytic solution with the stated strict bound. On WδW_\delta, restrict further to the Weierstrass domains max⁡i∣zi′−z0i∣≤γ\max_i|z'_i-z_{0i}|\leq\gamma, where 0<γ<10<\gamma<1. As δ\delta and γ\gamma vary, these domains cover WW admissibly. Their defining quotient norms make the map from AδA_\delta contractive and give ∥z′−z0∥≤γ\|z'-z_0\|\leq\gamma; thus all the restricted series substitutions converge. The same difference identity has an invertible matrix factor, since its error has norm at most max⁡(δ,γ)<1\max(\delta,\gamma)<1. It follows in each domain algebra that z′=zδz'=z_\delta, and sheaf gluing proves uniqueness on WW. ◻

Corollary 4.11.2. NAG-002034 (Pointed Jacobian charts preserve entire tubes). Let KK be nontrivially valued, put O=K∘O=K^\circ and k=K~k=\widetilde K, and let X=Spf⁡A0\mathfrak X=\operatorname{Spf}A_0 and Y=Spf⁡B0\mathfrak Y=\operatorname{Spf}B_0 be admissible formal OO-schemes with F=(F1,…,Fn)∈A0⟨Z1,…,Zn⟩n,J(Z)=(∂Fi/∂Zj),B0=A0⟨Z1,…,Zn,Q⟩/(F1,…,Fn,Qdet⁡J(Z)−1).\begin{gathered} F=(F_1,\ldots,F_n)\in A_0\langle Z_1,\ldots,Z_n\rangle^n, \qquad J(Z)=(\partial F_i/\partial Z_j),\\ B_0=A_0\langle Z_1,\ldots,Z_n,Q\rangle/ (F_1,\ldots,F_n,Q\det J(Z)-1). \end{gathered} Let q∈Xs(k)q\in\mathfrak X_s(k) and z0∈Onz_0\in O^n satisfy F‾(z‾0)(q)=0\overline F(\overline z_0)(q)=0 and det⁡J‾(z‾0)(q)≠0\det\overline J(\overline z_0)(q)\ne 0. Let q′∈Ys(k)q'\in\mathfrak Y_s(k) be the point over qq with Z(q′)=z‾0Z(q')=\overline z_0 and Q(q′)=det⁡J‾(z‾0)(q)−1Q(q')=\det\overline J(\overline z_0)(q)^{-1}. The generic-fibre map induces a canonical analytic isomorphism ]q′[Y → ∼  ]q[X.]q'[_{\mathfrak Y}\ \xrightarrow{\ \sim\ }\ ]q[_{\mathfrak X}. The same conclusion holds for a finite composition of such pointed charts and for formal-open restrictions containing the chosen special-fibre points.

Proof. Proposition 3.14.7 gives the generic affinoid algebras and their quotient norms. The coefficients of each FiF_i have norm at most one and tend to zero in this norm, because they lie in A0A_0 and tend to zero π\pi-adically. By Proposition 3.20.1, a point of ]q[X]q[_{\mathfrak X} satisfies max⁡i∣Fi(z0)∣<1\max_i|F_i(z_0)|<1 and ∣det⁡J(z0)∣=1|\det J(z_0)|=1. This tube therefore lies in WW.

Over this base tube, specialization to q′q' is equivalent to max⁡i∣Zi−z0i∣<1\max_i|Z_i-z_{0i}|<1. The base coordinates already specialize to qq, these inequalities fix the residues of all ZiZ_i, and Qdet⁡J(Z)=1Q\det J(Z)=1 forces the specified residue of QQ. These coordinates generate the special-fibre algebra and hence determine its specialization point. Thus ]q′[Y=Y∩p−1(]q[X).]q'[_{\mathfrak Y}=Y\cap p^{-1}(]q[_{\mathfrak X}). Restricting the preceding isomorphism proves the assertion. Two lifts of z‾0\overline z_0 differ coordinatewise by elements of K∘∘K^{\circ\circ} and define the same strict inequalities; the unique inverse is consequently independent of the lift. A formal open containing qq contains its entire tube in the generic fibre. Indeed, the principal-open specialization criterion applies to a principal neighborhood of qq inside that formal open. Formal-open restrictions therefore preserve the isomorphism, and composition gives the final assertion. ◻

4.12 Reduced models and component points

Let KK be complete and nontrivially valued, with its original absolute value. Put O=K∘O=K^\circ, m=K∘∘\mathfrak m=K^{\circ\circ}, and k=K~k=\widetilde K. Fix 0<∣π∣<10<|\pi|<1.

Lemma 4.12.1. NAG-002035 (A reduced special fibre determines the unit ball). Let A0A_0 be an admissible OO-algebra, let A=A0[1/π]A=A_0[1/\pi], and assume that A0/mA0A_0/\mathfrak mA_0 is reduced. Then the quotient norm associated to any admissible integral presentation of A0A_0 is power-multiplicative and equals the spectral radius. Moreover, A∘=A0,A∘∘=mA0,A~=A0/mA0.A^\circ=A_0,\qquad A^{\circ\circ}=\mathfrak mA_0, \qquad \widetilde A=A_0/\mathfrak mA_0. In particular, both AA and A0A_0 are reduced.

Proof. Use Proposition 3.14.7, including the norm-attaining lifts in its proof. It identifies A0A_0 with the quotient closed unit ball. For nonzero a∈Aa\in A, its norm is attained by a restricted power-series lift, whose nonzero norm is the maximum of its coefficient absolute values. Thus ∥a∥∈∣K×∣\|a\|\in|K^\times|. It follows that mA0={a∈A:∥a∥<1}.\mathfrak mA_0=\{a\in A:\|a\|<1\}. One inclusion follows from the ultrametric inequality for finite sums. For the other, choose c∈Kc\in K with ∣c∣=∥a∥<1|c|=\|a\|<1; then a/c∈A0a/c\in A_0. The case a=0a=0 is immediate.

For any nonzero aa, choose c∈K×c\in K^\times with ∣c∣=∥a∥|c|=\|a\| and put b=a/cb=a/c. Its class in A0/mA0A_0/\mathfrak mA_0 is nonzero. Reducedness implies that every positive power of this class is nonzero. Since ∥bj∥≤1\|b^j\|\leq 1, the displayed characterization of mA0\mathfrak mA_0 gives ∥bj∥=1\|b^j\|=1 for all j≥1j\geq 1. Consequently ∥aj∥=∥a∥j,ρA(a)=lim⁡j→∞∥aj∥1/j=∥a∥.\|a^j\|=\|a\|^j,\qquad \rho_A(a)=\lim_{j\to\infty}\|a^j\|^{1/j}=\|a\|. A nonzero nilpotent is incompatible with this identity. Power-bounded elements are exactly those of norm at most one, and topologically nilpotent elements are exactly those of norm less than one. This proves all three identifications. ◻

Proposition 4.12.2. NAG-002036 (The analytic point of a residual component). Let X\mathfrak X be an admissible formal OO-scheme with reduced special fibre, and let X=XηX=\mathfrak X_\eta. For each generic point η\eta of an irreducible component of Xs\mathfrak X_s, the fibre sp⁡−1(η)\operatorname{sp}^{-1}(\eta) consists of a unique point xη∈Xx_\eta\in X. On each formal affine open, these points are exactly the Shilov boundary of its generic fibre. They agree under restriction to formal opens. If X\mathfrak X is quasi-compact, the resulting component-point set is finite.

Proof. Choose a formal affine open U=Spf⁡A0\mathfrak U=\operatorname{Spf}A_0 containing η\eta. The preceding lemma identifies its special fibre with strict canonical reduction. Theorem 3.16.7 therefore gives a unique point above each component generic point and identifies these points with the Shilov boundary of Uη\mathfrak U_\eta.

The specialization and formal-open compatibility in Proposition 3.17.2 give sp⁡−1(Us)=Uη\operatorname{sp}^{-1}(\mathfrak U_s)=\mathfrak U_\eta. Thus every global point above η\eta already belongs to this affine generic fibre. This proves global uniqueness as well as existence. On two formal affine opens, restriction to an affine open in their intersection containing η\eta shows that the constructed points agree.

Finally, a finite formal affine cover exists when X\mathfrak X is quasi-compact. By Corollary 3.12.5, the generic fibre of each member has a finite Shilov boundary. Every global component generic point lies in a member of the cover, so its analytic lift belongs to one of these finitely many finite boundaries. This proves finiteness. ◻

4.13 Algebraic models and formal completion

Let KK be a complete nontrivially valued non-Archimedean field with its specified real-valued absolute value. Put O=K∘O=K^\circ and fix π∈O\pi\in O with 0<∣π∣<10<|\pi|<1. All completions below are π\pi-adic. The value group need not be discrete.

Lemma 4.13.1. NAG-002037 (Affine algebraic models). Let BB be a finitely presented flat OO-algebra, with a presentation B=O[T1,…,Tn]/(f1,…,fs).B=O[T_1,\ldots,T_n]/(f_1,\ldots,f_s). Its completion B^\widehat B is an admissible OO-algebra, and the natural maps give B^=O⟨T1,…,Tn⟩/(f1,…,fs),B^[1/π]=K⟨T1,…,Tn⟩/(f1,…,fs),B/πmB≃B^/πmB^(m≥1).\begin{aligned} \widehat B&=O\langle T_1,\ldots,T_n\rangle/(f_1,\ldots,f_s),\\ \widehat B[1/\pi] &=K\langle T_1,\ldots,T_n\rangle/(f_1,\ldots,f_s),\\ B/\pi^m B&\simeq\widehat B/\pi^m\widehat B\qquad(m\geq1). \end{aligned} The generic fibre of Spf⁡B^\operatorname{Spf}\widehat B is canonically the analytic domain ∣Ti∣≤1|T_i|\leq 1 in (Spec⁡B[1/π])an(\operatorname{Spec}B[1/\pi])^{\mathrm{an}}, with the displayed affinoid algebra. For b∈Bb\in B, the completion of B[1/b]B[1/b] is the principal formal-open algebra B[1/b]^≃B^⟨U⟩/(bU−1).\widehat{B[1/b]} \simeq\widehat B\langle U\rangle/(bU-1). Its generic fibre is the domain ∣b∣=1|b|=1 in the preceding bounded-coordinate domain. These identifications respect the polynomial coordinate maps and their restrictions.

Proof. The completion of O[T]O[T] is P=O⟨T⟩P=O\langle T\rangle: a compatible polynomial system modulo πm\pi^m gives one coefficient in the complete ring OO at each monomial, and for every mm only finitely many coefficients survive modulo πm\pi^m. Conversely such a restricted coefficient family gives the compatible polynomial system. The powers of π\pi define the given topology on OO.

Apply Lemma 3.14.8 to O[T]↠BO[T]\twoheadrightarrow B. It makes P→B^P\to\widehat B surjective, with kernel the closure of (f1,…,fs)(f_1,\ldots,f_s) from O[T]O[T]. The ideal (f1,…,fs)P(f_1,\ldots,f_s)P is closed by Proposition 3.14.5. Polynomial approximation of the finitely many coefficients in a sum ∑ifigi\sum_i f_i g_i shows that this ideal is exactly the stated closure. This proves the restricted quotient presentation.

Flatness of BB preserves injectivity of multiplication by π\pi on OO. Hence BB has no π\pi-torsion. Lemma 3.14.8 gives the same assertion for B^\widehat B and gives all the reduction isomorphisms. The displayed quotient is topologically of finite type; Theorem 3.14.4 makes it topologically finitely presented, so it is admissible. By Proposition 3.14.7, inverting π\pi gives the displayed strict affinoid algebra. Theorem 2.33.23, with all radii equal to one, identifies it with the asserted analytic domain, including its analytic structure.

Localization preserves OO-flatness. Apply the quotient calculation to B[1/b]=B[U]/(bU−1)B[1/b]=B[U]/(bU-1) to obtain the last presentation. Equivalently, its reductions modulo every πm\pi^m are (B/πmB)[1/b](B/\pi^mB)[1/b], so it is the principal open restriction of Spf⁡B^\operatorname{Spf}\widehat B. Proposition 3.17.1 identifies its generic domain by ∣b∣≥1|b|\geq 1; every b∈B^b\in\widehat B already has absolute value at most one there, giving ∣b∣=1|b|=1. All these maps agree on polynomial coordinates and their inverses, hence agree on the completions by continuity. ◻

Theorem 4.13.2. NAG-002038 (Completion of a projective model). Let C\mathcal C be a projective, finitely presented, flat OO-scheme, and let C^\widehat{\mathcal C} be its π\pi-adic formal completion. Then C^\widehat{\mathcal C} is an admissible formal model of (CK)an(\mathcal C_K)^{\mathrm{an}}: there is a canonical analytic isomorphism (C^)η≃(CK)an.(\widehat{\mathcal C})_\eta \simeq(\mathcal C_K)^{\mathrm{an}}. For a closed immersion C↪PON\mathcal C\hookrightarrow\mathbf P^N_O, the generic fibres of the completed standard affine charts are the domains Vj={x∈(CK)an:∣Xi/Xj∣(x)≤1 for every i≠j}.V_j=\{x\in(\mathcal C_K)^{\mathrm{an}}: |X_i/X_j|(x)\leq1\text{ for every }i\ne j\}. Their overlap maps are the usual coordinate-ratio maps.

Proof. Choose the indicated projective embedding and write Uj=C∩D+(Xj)=Spec⁡Bj\mathcal U_j=\mathcal C\cap D_+(X_j)=\operatorname{Spec}B_j. These are finitely presented flat affine OO-schemes: flatness and local finite presentation restrict to open subschemes, and each Uj\mathcal U_j is affine and quasi-compact. Its coordinate algebra is generated by the ratios Ti=Xi/XjT_i=X_i/X_j. A finite-type presentation using these chosen generators has finitely generated kernel. Indeed, if Bj=O[Z]/(F)B_j=O[Z]/(F) is any finite presentation, express Z=p(T)Z=p(T) and T=q(Z)T=q(Z) in BjB_j; the finitely many relations F(p(T))F(p(T)) and T−q(p(T))T-q(p(T)) generate the kernel, by substituting in a polynomial relation.

By Lemma 4.13.1, each B^j\widehat B_j is admissible. On Uj∩Uℓ=D(Xℓ/Xj)\mathcal U_j\cap\mathcal U_\ell=D(X_\ell/X_j), the same lemma identifies the completed overlap with the principal formal open of Spf⁡B^j\operatorname{Spf}\widehat B_j. The reduction formula modulo every πm\pi^m identifies these overlaps with the restrictions of C×OO/πm\mathcal C\times_O O/\pi^m. Their inverse limits therefore glue to exactly C^\widehat{\mathcal C}, proving its admissibility.

The generic fibre of the jjth completed chart is, by that lemma and Theorem 2.33.26, the actual closed-unit-coordinate domain VjV_j of (CK)an(\mathcal C_K)^{\mathrm{an}}, with the same quotient ideal. No reduction of that ideal is taken. The completed overlap adds the bound ∣Xj/Xℓ∣≤1|X_j/X_\ell|\leq 1 to ∣Xℓ/Xj∣≤1|X_\ell/X_j|\leq 1, hence gives the domain ∣Xℓ/Xj∣=1|X_\ell/X_j|=1 inside VjV_j. This is exactly Vj∩VℓV_j\cap V_\ell; the remaining coordinates change by division by Xℓ/XjX_\ell/X_j. Thus the coordinate identifications agree as bounded affinoid algebra homomorphisms on every overlap.

The domains VjV_j cover (CK)an(\mathcal C_K)^{\mathrm{an}}: at any analytic point choose a homogeneous coordinate of maximal absolute value. This is precisely the projective affinoid covering in Proposition 2.33.30. Proposition 3.17.2 constructs the generic fibre by gluing the completed affine charts along the overlaps just identified. The same domains and the same transition maps consequently give the asserted analytic isomorphism. On every chart its map is induced by Bj→B^j[1/π]B_j\to\widehat B_j[1/\pi], so these local maps define the natural comparison map to (CK)an(\mathcal C_K)^{\mathrm{an}}. Its definition on affine restrictions is independent of the projective embedding, which proves canonicity. ◻

4.14 Stable models over valuation rings

A stable family of genus g≥2g\geq 2 is a proper, flat, finitely presented family of curves whose geometric fibres are connected nodal curves of genus gg with ample dualizing sheaf. This is the stable-family convention of [StacksAlg] (Tags 0E75, 0E76, 0E78). A stable model of a smooth projective curve C/KC/K over a valuation ring R⊂KR\subset K includes an identification of its generic fibre with CC. We first obtain a finite extension of the fraction field by taking a closed point on a proper algebraic-space cover of the moduli stack.

Lemma 4.14.1. NAG-002043 (Finite extensions in the valuative criterion). Let RR be a valuation ring with fraction field KK, let X\mathcal X be an algebraic stack proper over Spec⁡R\operatorname{Spec}R, and let x:Spec⁡K→Xx:\operatorname{Spec}K\to\mathcal X be an RR-morphism. There is a finite field extension L/KL/K such that, for every valuation ring R′⊂LR'\subset L dominating RR and having fraction field LL, the object xLx_L extends to an object of X(R′)\mathcal X(R'). Such valuation rings R′R' exist.

Proof. The scheme Spec⁡R\operatorname{Spec}R is quasi-compact and quasi-separated, and X→Spec⁡R\mathcal X\to\operatorname{Spec}R is separated and of finite type. Chow’s lemma for algebraic stacks, [StacksAlg] (Tag 0CQ8), gives an algebraic space PP and a proper surjective morphism P→XP\to\mathcal X. Proper morphisms compose by [StacksAlg] (Tag 0CL7); hence P→Spec⁡RP\to\operatorname{Spec}R is proper. The compatibility of properness for algebraic spaces and stacks is recorded in [StacksAlg] (Tag 0CL4).

The fibre product Px=P×XSpec⁡KP_x=P\times_{\mathcal X}\operatorname{Spec}K is an algebraic space. Indeed, its map to P×RSpec⁡KP\times_R\operatorname{Spec}K is the base change of the representable diagonal of X\mathcal X; see [StacksAlg] (Tag 04XS). Properness and surjectivity survive this base change, so PxP_x is a nonempty proper algebraic space of finite type over KK. Choose a nonempty affine open W=Spec⁡BW=\operatorname{Spec}B in an étale scheme presentation of PxP_x. The morphism W→Spec⁡KW\to\operatorname{Spec}K is locally of finite type and quasi-compact, so BB is a finite-type KK-algebra. A maximal ideal of BB has residue field LL finite over KK by [StacksAlg] (Tag 00FV). The resulting point of Px(L)P_x(L) gives a point pL:Spec⁡L→Pp_L:\operatorname{Spec}L\to P and a specified isomorphism between its image in X(L)\mathcal X(L) and xLx_L.

Fix any valuation ring R′⊂LR'\subset L as in the statement. Apply the ordinary valuative criterion for the proper algebraic space P/RP/R, [StacksAlg] (Tag 0A40), to Spec⁡L⟶P,Spec⁡R′⟶Spec⁡R.\operatorname{Spec}L\longrightarrow P, \qquad \operatorname{Spec}R'\longrightarrow\operatorname{Spec}R. It supplies a unique extension Spec⁡R′→P\operatorname{Spec}R'\to P of pLp_L. Its composite with P→XP\to\mathcal X, together with the preceding generic isomorphism, is the required extension of xLx_L. Finally, [StacksAlg] (Tag 00IA) applied to the local subring R⊂LR\subset L supplies such an R′R'. Domination ensures R′∩K=RR'\cap K=R: if a∈K∖Ra\in K\setminus R, then a−1∈mR⊂mR′a^{-1}\in\mathfrak m_R\subset\mathfrak m_{R'}, which excludes a∈R′a\in R'. ◻

Theorem 4.14.2. NAG-002044 (Stable reduction over a valuation ring). Let RR be a valuation ring with fraction field KK, and let C/KC/K be a smooth projective geometrically connected curve of genus g≥2g\geq 2. There is a finite field extension L/KL/K such that, for every valuation ring R′⊂LR'\subset L with fraction field LL extending RR, the curve CLC_L has a stable model C/R′\mathcal C/R' which is a projective scheme, flat and of finite presentation over R′R'. If RR has rank one, then R′R' has rank one.

Proof. The curve CC gives a KK-object of the stack M‾g,R\overline{\mathcal M}_{g,R} of stable curves of genus gg, by [StacksAlg] (Tags 0E75, 0E78). This stack is proper over RR: [StacksAlg] (Tag 0E9C) proves properness over Z\mathbf Z, and [StacksAlg] (Tag 0CL6) preserves it under base change. Lemma 4.14.1 therefore gives a finite extension L/KL/K and, over each R′R' as stated, a stable family of algebraic spaces f:C→Spec⁡R′f:\mathcal C\to\operatorname{Spec}R' whose generic fibre is identified with CLC_L. As an object of the stack of curves, this family is proper, flat and of finite presentation by [StacksAlg] (Tag 0D4Z).

We verify that this family is a projective scheme. Let ss be the closed point of Spec⁡R′\operatorname{Spec}R'. Stability of Cs\mathcal C_s and [StacksAlg] (Tag 0E74) give an open neighbourhood UU of ss on which ωC/R′\omega_{\mathcal C/R'} is relatively ample. Every open neighbourhood of the closed point of a local spectrum is the whole spectrum: a basic open D(a)D(a) containing that point has aa a unit. Hence U=Spec⁡R′U=\operatorname{Spec}R'. By [StacksAlg] (Tag 0D32), relative ampleness over the scheme Spec⁡R′\operatorname{Spec}R' implies that C\mathcal C is a scheme. Properness and the ample invertible sheaf make ff locally projective by [StacksAlg] (Tag 0B5N). Apply the same local-spectrum observation to the projective neighbourhood of ss; thus ff is projective.

For the last assertion, the value group of RR has finite index in that of R′R' by [StacksAlg] (Tag 0ASH). Write the groups additively as Γ⊂Γ′\Gamma\subset\Gamma'. Every element of Γ′\Gamma' has a positive integer multiple in Γ\Gamma. If H′⊂Γ′H'\subset\Gamma' is a nonzero convex subgroup, multiplying one of its nonzero elements shows that H′∩ΓH'\cap\Gamma is nonzero and convex in Γ\Gamma. Rank one of Γ\Gamma gives H′∩Γ=ΓH'\cap\Gamma=\Gamma. For any positive γ′∈Γ′\gamma'\in\Gamma', some nγ′n\gamma' lies in Γ⊂H′\Gamma\subset H'; convexity then gives γ′∈H′\gamma'\in H'. Thus H′=Γ′H'=\Gamma', proving that Γ′\Gamma' has rank one. ◻

Corollary 4.14.3. NAG-002045 (Stable models for the fixed absolute value). Let KK be a complete non-Archimedean field with its given nontrivial absolute value, and let C/KC/K be a smooth projective geometrically connected curve of genus at least two. There is a finite extension L/KL/K, endowed with the unique extending absolute value, for which CLC_L has a projective stable model over L∘L^\circ. If char⁡K=0\operatorname{char}K=0, the extension L/KL/K may be chosen Galois.

Proof. Apply Theorem 4.14.2 to R=K∘R=K^\circ. Lemma 1.12.1 supplies the unique extending absolute value on the finite extension LL; its valuation ring L∘L^\circ dominates K∘K^\circ, and LL is complete. We may therefore take R′=L∘R'=L^\circ in that theorem.

If char⁡K=0\operatorname{char}K=0, take a finite normal closure M/KM/K of L/KL/K using [StacksAlg] (Tag 09DT). Every irreducible polynomial over a characteristic-zero field has nonzero derivative and is relatively prime to its derivative, so M/KM/K is separable and hence Galois. Uniqueness of the extended absolute values identifies the restriction from MM with the chosen value on LL, giving L∘⊂M∘L^\circ\subset M^\circ. Base change of the stable model to M∘M^\circ remains stable by [StacksAlg] (Tag 0E76), and remains projective, flat and of finite presentation. Replacing LL by MM proves the assertion. ◻

Lemma 4.14.4. NAG-002046 (Extension of an isomorphism of stable generic fibres). Let RR be a valuation ring with fraction field KK, and let C1/R\mathcal C_1/R and C2/R\mathcal C_2/R be stable families of the same genus g≥2g\geq 2. Every isomorphism (C1)K≃(C2)K(\mathcal C_1)_K\simeq(\mathcal C_2)_K extends uniquely to an RR-isomorphism C1≃C2\mathcal C_1\simeq\mathcal C_2.

Proof. The moduli stack M‾g\overline{\mathcal M}_g is separated by [StacksAlg] (Tag 0E9A). By the definition of separatedness for stacks, [StacksAlg] (Tag 04YW), its diagonal is proper. Pulling this diagonal back by the pair (C1,C2)(\mathcal C_1,\mathcal C_2) gives the algebraic space I=Isom⁡R(C1,C2)⟶Spec⁡R;I=\operatorname{Isom}_R(\mathcal C_1,\mathcal C_2) \longrightarrow\operatorname{Spec}R; the identification with the isomorphism sheaf is the fibre-product description in [StacksAlg] (Tag 04XS). The morphism I→Spec⁡RI\to\operatorname{Spec}R is proper by base change. A specified generic isomorphism is a KK-point of II. The proper algebraic-space valuative criterion, [StacksAlg] (Tag 0A40), extends it uniquely to an RR-point of II, which is precisely the required isomorphism. ◻

Corollary 4.14.5. NAG-002047 (The semilinear action on a stable model). Let KK be a complete non-Archimedean field, let L/KL/K be finite Galois with group GG, and give LL its unique extending absolute value. Let C/KC/K be a smooth projective geometrically connected curve of genus g≥2g\geq 2, and let C/L∘\mathcal C/L^\circ be a stable model of CLC_L with its generic-fibre identification. The canonical semilinear GG-action on CLC_L extends uniquely to a semilinear GG-action on C\mathcal C. More explicitly, with S=Spec⁡L∘S=\operatorname{Spec}L^\circ and bσ=Spec⁡(σ−1∣L∘)b_\sigma=\operatorname{Spec}(\sigma^{-1}|_{L^\circ}), there are unique automorphisms ρσ:C→C\rho_\sigma:\mathcal C\to\mathcal C satisfying f∘ρσ=bσ∘f,ρ1=id⁡,ρστ=ρσ∘ρτ,f\circ\rho_\sigma=b_\sigma\circ f, \qquad \rho_1=\operatorname{id}, \qquad \rho_{\sigma\tau}=\rho_\sigma\circ\rho_\tau, and inducing the canonical action on the generic fibre.

Proof. Uniqueness in Lemma 1.12.1 gives ∣σ(a)∣=∣a∣|\sigma(a)|=|a| for every a∈La\in L and σ∈G\sigma\in G, so GG preserves L∘L^\circ. Our inverse convention gives bσbτ=bστb_\sigma b_\tau=b_{\sigma\tau}. The scheme CL=C×KSpec⁡LC_L=C\times_K\operatorname{Spec}L has the corresponding canonical left action induced by id⁡C×Spec⁡(σ−1)\operatorname{id}_C\times\operatorname{Spec}(\sigma^{-1}).

For a fixed σ\sigma, form bσ∗C=C×S,bσSb_\sigma^*\mathcal C=\mathcal C\times_{S,b_\sigma}S, with its structure map given by the second projection. Both this twist and C\mathcal C are stable families over SS. The generic action gives an LL-isomorphism CL→(bσ∗C)L\mathcal C_L\to(b_\sigma^*\mathcal C)_L: its two components are the generic semilinear automorphism and the structure map to Spec⁡L\operatorname{Spec}L. Lemma 4.14.4 extends it uniquely to an SS-isomorphism aσ:C→bσ∗Ca_\sigma:\mathcal C\to b_\sigma^*\mathcal C. Composing with the first projection defines ρσ\rho_\sigma. Since bσb_\sigma is an automorphism, that projection is an isomorphism of schemes, and ρσ\rho_\sigma is an automorphism covering bσb_\sigma.

The maps ρσρτ\rho_\sigma\rho_\tau and ρστ\rho_{\sigma\tau} cover the same base map and agree on the generic fibre. Reinterpreting them as SS-isomorphisms C→bστ∗C\mathcal C\to b_{\sigma\tau}^*\mathcal C, uniqueness in Lemma 4.14.4 makes them equal. The same argument gives ρ1=id⁡\rho_1=\operatorname{id} and proves uniqueness of the action. ◻

4.15 Pointed nodal charts over the valuation ring

Let KK be a complete non-Archimedean field with its given nontrivial absolute value. Write O=K∘O=K^\circ, m=K∘∘\mathfrak m=K^{\circ\circ} and k=O/mk=O/\mathfrak m. We turn the étale base neighbourhood in the nodal normal-form theorem into the spectrum of the valuation ring of a finite extension. The two field-theoretic ingredients are simple-root lifting and a lift of a finite separable residue extension.

Lemma 4.15.1. NAG-002049 (Simple roots over a complete valued field). The valuation ring OO is henselian. If V→Spec⁡OV\to\operatorname{Spec}O is étale and v∈Vv\in V lies over the closed point with κ(v)=k\kappa(v)=k, then an affine open neighbourhood of vv admits a section over Spec⁡O\operatorname{Spec}O through vv. Its image is open in VV.

Proof. Let F∈O[T]F\in O[T] have a simple root b‾∈k\overline b\in k after reduction, and choose a lift b∈Ob\in O. Then ∣F(b)∣<1|F(b)|<1 and ∣F′(b)∣=1|F'(b)|=1. Expanding at bb gives F(b+Z)F′(b)=c0+Z+∑i=2dciZi,∣c0∣<1,∣ci∣≤1.\frac{F(b+Z)}{F'(b)}=c_0+Z+\sum_{i=2}^d c_iZ^i, \qquad |c_0|<1,\quad |c_i|\leq1. Choose a real number δ\delta with ∣c0∣<δ<1|c_0|<\delta<1. On the complete closed ball ∣Z∣≤δ|Z|\leq\delta, the map Φ(Z)=−c0−∑i≥2ciZi\Phi(Z)=-c_0-\sum_{i\geq 2}c_iZ^i takes values in the same ball and satisfies ∣Φ(Z)−Φ(Z′)∣≤δ ∣Z−Z′∣.|\Phi(Z)-\Phi(Z')|\leq\delta\,|Z-Z'|. Starting with Z0=0Z_0=0, its iterates satisfy ∣Zn+1−Zn∣≤δn∣Z1∣|Z_{n+1}-Z_n|\leq\delta^n|Z_1|; the ultrametric inequality makes them Cauchy. Their limit ZZ is a fixed point. Thus b+Z∈Ob+Z\in O lifts b‾\overline b and satisfies F(b+Z)=0F(b+Z)=0. This is the one-variable contraction used in the proof of Lemma 2.54.1, now applied to a residue root. The simple-root characterization in [StacksAlg] (Tag 04GG) proves that OO is henselian.

Choose an affine open V0=Spec⁡BV_0=\operatorname{Spec}B containing vv. The map O→BO\to B is étale. Part (8) of [StacksAlg] (Tag 04GG) supplies the OO-algebra retraction B→OB\to O whose inverse image of m\mathfrak m is the prime of vv. The corresponding section passes through vv. A section of an unramified morphism is an open immersion by [StacksAlg] (Tag 024T); hence its image is open in V0V_0 and in VV. ◻

Lemma 4.15.2. NAG-002050 (Lifting a finite separable residue extension). Let l/kl/k be a finite separable extension. There is a finite separable extension L/KL/K, with the unique extending absolute value, and a kk-isomorphism l≃L~l\simeq\widetilde L such that [L:K]=[L~:k]=[l:k],∣L×∣=∣K×∣.[L:K]=[\widetilde L:k]=[l:k], \qquad |L^\times|=|K^\times|.

Proof. By the primitive-element theorem, [StacksAlg] (Tag 030N), write l=k(β)l=k(\beta) with monic irreducible polynomial p(T)∈k[T]p(T)\in k[T] of degree d=[l:k]d=[l:k]. Lift its coefficients to a monic polynomial P(T)∈O[T]P(T)\in O[T] of degree dd. Choose a root α\alpha of PP in an algebraic extension of KK, set L=K(α)L=K(\alpha), and use Lemma 1.12.1 to equip LL with its complete extending absolute value. Then [L:K]≤d[L:K]\leq d.

If ∣α∣>1|\alpha|>1, the leading term of P(α)P(\alpha) strictly dominates every other term, which is impossible. Thus α∈L∘\alpha\in L^\circ, and p(α‾)=0p(\overline\alpha)=0 defines an embedding l↪L~l\hookrightarrow\widetilde L sending β\beta to α‾\overline\alpha. The value-group and residue-degree inequality [StacksAlg] (Tag 0ASH) gives d≤[L~:k]≤[∣L×∣:∣K×∣] [L~:k]≤[L:K]≤d.d\leq[\widetilde L:k]\leq [|L^\times|:|K^\times|]\,[\widetilde L:k] \leq[L:K]\leq d. All inequalities are equalities. In particular the residue embedding is an isomorphism, the value-group index is one, and PP is the minimal polynomial of α\alpha. Since pp is separable, p′(α‾)≠0p'(\overline\alpha)\ne 0, so P′(α)P'(\alpha) is a unit of L∘L^\circ. The minimal polynomial is therefore separable. ◻

Theorem 4.15.3. NAG-002051 (A rational pointed nodal chart after finite extension). Let X→Spec⁡O\mathcal X\to\operatorname{Spec}O be a morphism of schemes which is at-worst-nodal of relative dimension one, and let xx be a node of its closed fibre. There are a finite separable extension L/KL/K, an affine L∘L^\circ-scheme UU, étale morphisms XL∘ ⟵ U ⟶ Wa=Spec⁡(L∘[s,t]/(st−a))(a∈L∘∘),\mathcal X_{L^\circ}\ \longleftarrow\ U \ \longrightarrow\ W_a=\operatorname{Spec}\bigl(L^\circ[s,t]/(st-a)\bigr) \qquad(a\in L^{\circ\circ}), and a L~\widetilde L-rational point uu of the closed fibre of UU such that its image x′x' in XL~\mathcal X_{\widetilde L} lies over xx and its image in (Wa)L~(W_a)_{\widetilde L} is the origin s=t=0s=t=0. The point x′x' is a split node, with O^XL~,x′≃L~[[s,t]]/(st).\widehat{\mathcal O}_{\mathcal X_{\widetilde L},x'} \simeq \widetilde L[[s,t]]/(st). If the generic fibre XK\mathcal X_K is smooth over KK, the parameter satisfies a≠0a\ne 0, and consequently 0<∣a∣<10<|a|<1.

Proof. Apply the nodal normal-form theorem [StacksAlg] (Tag 0CBY). It gives étale maps X←U0\mathcal X\leftarrow U_0, V→Spec⁡OV\to\operatorname{Spec}O and U0⟶V×Spec⁡Z[A]Spec⁡(Z[S,T,A]/(ST−A)),U_0\longrightarrow V\times_{\operatorname{Spec}\mathbf Z[A]} \operatorname{Spec}\bigl(\mathbf Z[S,T,A]/(ST-A)\bigr), and a point u0∈U0u_0\in U_0 mapping to xx. Let v0∈Vv_0\in V be its image. A nodal point of a curve has finite separable residue field over the ground field by [StacksAlg] (Tag 0C4D). The étale fibre description [StacksAlg] (Tag 02GL) makes κ(u0)/κ(x)\kappa(u_0)/\kappa(x) finite separable as well. Hence l=κ(u0)l=\kappa(u_0) is finite separable over kk.

Choose L/KL/K by Lemma 4.15.2, with a specified identification l=L~l=\widetilde L. Base change the diagram to O′=L∘O'=L^\circ. The maps Spec⁡L~→U0\operatorname{Spec}\widetilde L\to U_0 and Spec⁡L~→Spec⁡O′\operatorname{Spec}\widetilde L\to\operatorname{Spec}O' define a rational point u1u_1 of (U0)O′(U_0)_{O'}, whose image v1v_1 in VO′V_{O'} is rational. Replace VO′V_{O'} by an affine open containing v1v_1 and (U0)O′(U_0)_{O'} by its inverse image. Since LL is complete, Lemma 4.15.1 gives a section j:Spec⁡O′⟶VO′j:\operatorname{Spec}O'\longrightarrow V_{O'} through v1v_1. Pull back the diagram along jj. Its image is open, so the resulting U1U_1 is open in (U0)O′(U_0)_{O'}; in particular U1→XO′U_1\to\mathcal X_{O'} remains étale. The other map becomes étale over Wa=Spec⁡(O′[s,t]/(st−a))W_a=\operatorname{Spec}(O'[s,t]/(st-a)), where a∈O′a\in O' is the pullback of the parameter AA. These uses of composition and base change are [StacksAlg] (Tags 02GN, 02GO). The point u1u_1 belongs to this pullback because its image is the special point of the section. Choose an affine open neighbourhood U⊂U1U\subset U_1 of it and denote the point by uu.

Let x′x' and ww be the images of uu in XL~\mathcal X_{\widetilde L} and (Wa)L~(W_a)_{\widetilde L}. Both are rational: their residue fields are intermediate fields between L~\widetilde L and κ(u)=L~\kappa(u)=\widetilde L. The point x′x' remains a node by the base-change invariance in [StacksAlg] (Tag 0C56). Applying the étale invariance of nodes, [StacksAlg] (Tag 0C57), to the two maps from UL~U_{\widetilde L} shows that uu and ww are nodes. The curve st−a‾=0st-\overline a=0 is smooth wherever ss or tt is invertible; its only possible nodal point is therefore s=t=0s=t=0. Thus ww is that origin and a‾=0\overline a=0.

The local rings at x′x', uu and ww of these special fibres are Noetherian, since the fibres are locally of finite type over L~\widetilde L. Each of the two étale local maps is flat, has the same residue field L~\widetilde L, and carries the source maximal ideal onto the target maximal ideal; flatness and the fibre assertion follow from [StacksAlg] (Tags 02GS, 02GL). Thus [StacksAlg] (Tag 0AGX) identifies their completions. The completion at ww is L~[[s,t]]/(st)\widetilde L[[s,t]]/(st), giving the displayed isomorphism for x′x'. The split-node criterion [StacksAlg] (Tag 0CBW) completes this assertion.

Finally assume that XK\mathcal X_K is smooth, and suppose a=0a=0. Then WaW_a has the section s=t=0s=t=0 over Spec⁡O′\operatorname{Spec}O'. The image of U→WaU\to W_a is open by [StacksAlg] (Tag 03WT) and contains the closed point of this section. Its inverse image on Spec⁡O′\operatorname{Spec}O' is an open neighbourhood of the closed point, hence the whole local spectrum. In particular the image contains the generic point of the section, the node (s,t)=(0,0)(s,t)=(0,0) of (W0)L(W_0)_L. Any point of ULU_L mapping to this node is itself a node by [StacksAlg] (Tag 0C57). On the other hand UL→XLU_L\to\mathcal X_L is étale and XL\mathcal X_L is smooth, so ULU_L is smooth; preservation of smoothness under étale precomposition is included in [StacksAlg] (Tag 036U). This contradiction proves a≠0a\ne 0. Together with a∈L∘∘a\in L^{\circ\circ} it gives 0<∣a∣<10<|a|<1. ◻

4.16 Coordinate changes on annuli

Let KK be a complete non-Archimedean field, with its specified absolute value. We use the Gauss skeleton, radial retraction and strong homotopy of Proposition 4.3.1. The parameter of that homotopy is denoted by u∈[0,1]u\in[0,1].

Theorem 4.16.1. NAG-002030 (Annulus isomorphisms and radial homotopies). Let 0<r<R<∞0<r<R<\infty and 0<r′<R′<∞0<r'<R'<\infty, and let ϕ:X=AK[r,R]→∼X′=AK[r′,R′]\phi:X=A_K[r,R]\xrightarrow{\sim}X'=A_K[r',R'] be an analytic isomorphism. Write T,ST,S for the coordinates of X,X′X,X'. There are c∈K×c\in K^\times, ε∈{1,−1}\varepsilon\in\{1,-1\} and h∈O(X)h\in\mathcal O(X) with ∥h∥r,R<1\|h\|_{r,R}<1 such that ϕ∗S=cTε(1+h).\phi^*S=cT^\varepsilon(1+h). For every r≤ρ≤Rr\leq\rho\leq R, ϕ(ζ0,ρ)=ζ0,∣c∣ρε′.\phi(\zeta_{0,\rho})=\zeta'_{0,|c|\rho^\varepsilon}. Thus ϕ\phi restricts to an isometry of Gauss skeleta for their logarithmic metrics. In coordinates t=−log⁡∣T∣t=-\log|T| and t′=−log⁡∣S∣t'=-\log|S|, this isometry is t′=εt+v(c)t'=\varepsilon t+v(c), where v(c)=−log⁡∣c∣v(c)=-\log|c|. Moreover, for every x∈Xx\in X and 0≤u≤10\leq u\leq 1, ϕ(HX(x,u))=HX′(ϕ(x),u),ϕ∘τX=τX′∘ϕ.\phi\bigl(H_X(x,u)\bigr)=H_{X'}\bigl(\phi(x),u\bigr), \qquad \phi\circ\tau_X=\tau_{X'}\circ\phi.

Proof. Put g=ϕ∗Sg=\phi^*S. Proposition 4.10.2, applied to gg and to the inverse isomorphism, gives g=cTm(1+h),∥h∥r,R<1,(ϕ−1)∗T=dSn(1+k),∥k∥r′,R′<1,\begin{aligned} g&=cT^m(1+h),&\qquad&\|h\|_{r,R}<1,\\ (\phi^{-1})^*T&=dS^n(1+k),&&\|k\|_{r',R'}<1, \end{aligned} with c,d∈K×c,d\in K^\times and m,n∈Zm,n\in\mathbb Z. The norm identity in that proposition holds at every analytic point. Composing the two identities at a point of radius ρ\rho gives ρ=∣d∣ ∣c∣nρmn(r≤ρ≤R).\rho=|d|\,|c|^n\rho^{mn}\qquad(r\leq\rho\leq R). Comparing two distinct radii gives mn=1mn=1. Hence m=n=ε∈{1,−1}m=n=\varepsilon\in\{1,-1\}. Surjectivity of ϕ\phi and the existence of the Gauss point at every radius show that ρ↦∣c∣ρε\rho\mapsto|c|\rho^\varepsilon maps [r,R][r,R] onto [r′,R′][r',R'].

We identify the image of each Gauss point. Set q=∥h∥r,R<1q=\|h\|_{r,R}<1, σ=∣c∣ρε\sigma=|c|\rho^\varepsilon, and take a nonzero Laurent polynomial P(S)=∑jajSjP(S)=\sum_j a_jS^j. For every integer jj, the geometric inverse and the integer binomial formula give ∣(1+h)j−1∣ζ0,ρ≤q.\bigl|(1+h)^j-1\bigr|_{\zeta_{0,\rho}}\leq q. Consequently ∣P(g)−∑jajcjTεj∣ζ0,ρ≤qmax⁡j∣aj∣σj.\left|P(g)-\sum_j a_jc^jT^{\varepsilon j}\right|_{\zeta_{0,\rho}} \leq q\max_j|a_j|\sigma^j. The exponents εj\varepsilon j are distinct. The Gauss norm of the Laurent polynomial on the right inside the difference is therefore max⁡j∣aj∣σj>0\max_j|a_j|\sigma^j>0, while the error has strictly smaller norm. Thus ∣P(g)∣ζ0,ρ=max⁡j∣aj∣σj|P(g)|_{\zeta_{0,\rho}}=\max_j|a_j|\sigma^j. Laurent polynomials are dense in the target affinoid algebra, so bounded multiplicative seminorms agreeing on them agree on the entire algebra. This proves the asserted equality of actual analytic points.

For an arbitrary xx, the norm formula gives ∣S(ϕ(x))∣=∣c∣ ∣T(x)∣ε|S(\phi(x))|=|c|\,|T(x)|^\varepsilon. Combining this with the Gauss-point equality proves ϕτX=τX′ϕ\phi\tau_X=\tau_{X'}\phi. Taking negative logarithms gives the stated affine formula and its preservation of logarithmic distance.

It remains to compare the homotopies for 0<u<10<u<1. Work in F=H(x)F=\mathcal H(x) and put α=T(x),ρ=∣α∣,β=g(x),σ=∣β∣=∣c∣ρε,δ=uρ.\alpha=T(x),\quad \rho=|\alpha|,\quad \beta=g(x),\quad \sigma=|\beta|=|c|\rho^\varepsilon, \qquad \delta=u\rho. For a Laurent series h=∑n∈ZhnTnh=\sum_{n\in\mathbb Z}h_nT^n, its substitution on the disc ∣Z∣≤δ<ρ|Z|\leq\delta<\rho is h(α+Z)=∑j≥0ejZj,ej=∑n∈Zhn(nj)αn−j,∣ej∣ρj≤q.h(\alpha+Z)=\sum_{j\geq0}e_jZ^j, \qquad e_j=\sum_{n\in\mathbb Z}h_n\binom nj\alpha^{n-j}, \qquad |e_j|\rho^j\leq q. To check the convergence, the weighted terms are bounded by ∣hn∣ρnuj|h_n|\rho^n u^j. The first factor tends to zero as ∣n∣→∞|n|\to\infty, and uju^j tends to zero as j→∞j\to\infty. Above any fixed positive threshold only finitely many pairs remain. Lemma 1.1.2 therefore permits the indicated sums and their interchange. The negative powers use the convergent geometric inverse of α+Z\alpha+Z. All integer binomial coefficients have absolute value at most one, which gives the displayed bound.

Expand g=cTε(1+h)g=cT^\varepsilon(1+h) at α\alpha as g(α+Z)=β+∑j≥1bjZj.g(\alpha+Z)=\beta+\sum_{j\geq1}b_jZ^j. The binomial coefficients of (α+Z)ε(\alpha+Z)^\varepsilon have the same bound, so ∣bj∣ρj≤σ|b_j|\rho^j\leq\sigma for j≥1j\geq 1. The first coefficient satisfies b1=cαε−1(ε(1+e0)+αe1),∣b1∣=σρ.b_1=c\alpha^{\varepsilon-1} \bigl(\varepsilon(1+e_0)+\alpha e_1\bigr), \qquad |b_1|=\frac{\sigma}{\rho}. Indeed ε=±1\varepsilon=\pm 1 has absolute value one, and both e0e_0 and αe1\alpha e_1 have size at most q<1q<1. Hence, in the radius-δ\delta Tate algebra over FF, g(α+Z)−β=b1Z(1+E(Z)),E(Z)=∑j≥2bjb1Zj−1,∥E∥δ≤u<1.g(\alpha+Z)-\beta=b_1Z(1+E(Z)),\qquad E(Z)=\sum_{j\geq2}\frac{b_j}{b_1}Z^{j-1},\qquad \|E\|_\delta\leq u<1. The bound follows termwise from ∣bj/b1∣δj−1≤uj−1|b_j/b_1|\delta^{j-1}\leq u^{j-1}; the same estimate proves convergence.

For a nonzero polynomial Q(S)∈K[S]Q(S)\in K[S], write Q(β+W)=∑j=0NqjWjQ(\beta+W)=\sum_{j=0}^N q_jW^j over FF, and put M=max⁡j∣qj∣(uσ)j>0.M=\max_j|q_j|(u\sigma)^j>0. Replacing W=b1Z(1+E)W=b_1Z(1+E) by b1Zb_1Z changes the resulting polynomial by a series of radius-δ\delta norm at most uM<MuM<M. The Gauss norm of ∑jqjb1jZj\sum_jq_jb_1^jZ^j is MM. Consequently ∥Q(g(α+Z))∥δ=max⁡j∣qj∣(uσ)j.\bigl\|Q\bigl(g(\alpha+Z)\bigr)\bigr\|_\delta =\max_j|q_j|(u\sigma)^j. The left side is the evaluation of QQ at ϕ(HX(x,u))\phi(H_X(x,u)). This follows from the construction of HXH_X and the bounded substitution T↦α+ZT\mapsto\alpha+Z, whose coordinate and inverse have norms ρ\rho and ρ−1\rho^{-1}. The right side is the evaluation at HX′(ϕ(x),u)H_{X'}(\phi(x),u); the induced embedding H(ϕ(x))↪F\mathcal H(\phi(x))\hookrightarrow F is isometric and hence preserves the Gauss norm. Thus these two points agree on K[S]K[S]. Their common nonzero value on SS gives agreement on Laurent polynomials, and density gives agreement on O(X′)\mathcal O(X'). This proves the homotopy identity for 0<u<10<u<1. At u=0u=0 it is the identity map, and at u=1u=1 it is the already proved compatibility of the radial retractions. ◻

Corollary 4.16.2. NAG-002031 (The original absolute value and edge length). Isomorphic nondegenerate closed annuli over KK have equal logarithmic moduli. For the nodal chart Spf⁡K∘⟨U,V⟩/(UV−a),0<∣a∣<1,\operatorname{Spf}K^\circ\langle U,V\rangle/(UV-a), \qquad 0<|a|<1, the radial edge has length −log⁡∣a∣-\log|a|, and this length and the preceding coordinate-change identities are preserved by complete isometric extensions of KK.

Proof. The two logarithmic radius intervals are related by an affine map of slope ±1\pm 1, so log⁡(R/r)=log⁡(R′/r′)\log(R/r)=\log(R'/r'). Lemma 3.20.3 and Proposition 3.26.1, with r=d=1r=d=1 in the latter’s notation, identify the nodal generic fibre with AK[∣a∣,1]A_K[|a|,1]. Its length is therefore −log⁡∣a∣-\log|a|. A complete isometric field extension leaves the coefficients’ absolute values, the Laurent coefficient norms and ∣a∣|a| unchanged. Base change of an analytic isomorphism is an isomorphism by Theorem 2.40.12; reapplying the theorem over the extended field gives the asserted identities. ◻

4.17 Tubes in algebraic models

Let KK be complete and nontrivially valued, with O=K∘O=K^\circ, k=K~k=\widetilde K, and a fixed π∈O\pi\in O satisfying 0<∣π∣<10<|\pi|<1. For a flat finitely presented OO-scheme X\mathcal X, write X^\widehat{\mathcal X} for its π\pi-adic completion. All tubes below are taken in its formal generic fibre.

Proposition 4.17.1. NAG-002039 (An algebraic pointed étale map preserves the tube). Let f:Y→Xf:\mathcal Y\to\mathcal X be an étale morphism between flat finitely presented OO-schemes. Let q′∈Ys(k)q'\in\mathcal Y_s(k) and q=f(q′)∈Xs(k)q=f(q')\in\mathcal X_s(k). The induced generic-fibre morphism restricts to an analytic isomorphism ]q′[Y^ → ∼  ]q[X^.]q'[_{\widehat{\mathcal Y}} \ \xrightarrow{\ \sim\ }\ ]q[_{\widehat{\mathcal X}}.

Proof. Lemma 4.13.1 makes these completions admissible, and is compatible with formal open restriction. A Zariski open containing the selected residual point retains its entire tube. We may therefore replace the source and target by affine neighbourhoods of q′q' and qq.

By the square Jacobian presentation in [StacksAlg] (Tag 02GU), after such shrinking the morphism is given by R→SR\to S, where S=R[Z1,…,Zn]/(F1,…,Fn),J(Z)=(∂Fi∂Zj)i,j,det⁡J(q′)≠0.\begin{gathered} S=R[Z_1,\ldots,Z_n]/(F_1,\ldots,F_n),\\ J(Z)=\left(\frac{\partial F_i}{\partial Z_j}\right)_{i,j}, \qquad \det J(q')\ne0. \end{gathered} Localize once more at det⁡J\det J, which is nonzero at q′q'. Adjoining its inverse as a variable QQ gives the presentation (F1,…,Fn,Qdet⁡J−1)(F_1,\ldots,F_n,Q\det J-1). Completing the polynomial presentation by Lemma 4.13.1 gives S^=R^⟨Z1,…,Zn,Q⟩/(F1,…,Fn,Qdet⁡J−1).\widehat S=\widehat R\langle Z_1,\ldots,Z_n,Q\rangle/ (F_1,\ldots,F_n,Q\det J-1). Here quotients are the separated completed quotients; their indicated finite relation ideals are closed by the admissible-presentation result. Choose lifts z0i∈Oz_{0i}\in O of the coordinates Zi(q′)∈kZ_i(q')\in k. They satisfy the hypotheses of Corollary 4.11.2, which supplies the required entire-tube isomorphism, including its analytic structure. The case of no variables is the identity case. All the preceding restrictions preserved the tubes, completing the proof. ◻

Corollary 4.17.2. NAG-002040 (The tube of a rational smooth point). Let X\mathcal X be a flat finitely presented OO-scheme. Suppose that X→Spec⁡O\mathcal X\to\operatorname{Spec}O is smooth of relative dimension dd at q∈Xs(k)q\in\mathcal X_s(k). Then ]q[X^]q[_{\widehat{\mathcal X}} is analytically isomorphic to the open unit dd-dimensional polydisc over KK.

Proof. By [StacksAlg] (Tag 054L), a Zariski neighbourhood U\mathcal U of qq admits an étale morphism U→AOd\mathcal U\to\mathbb A^d_O. Its image aˉ\bar a is a kk-rational point. Apply the preceding proposition. The completion of AOd\mathbb A^d_O is Spf⁡O⟨T1,…,Td⟩\operatorname{Spf}O\langle T_1,\ldots,T_d\rangle; its generic fibre is the closed unit polydisc. If ai∈Oa_i\in O lift the coordinates of aˉ\bar a, the specialization criterion identifies its aˉ\bar a-tube with ∣Ti−ai∣<1|T_i-a_i|<1 for every ii. Translation by aa gives the stated open unit polydisc. ◻

Corollary 4.17.3. NAG-002041 (Transport of a split nodal chart). Let 0<∣a∣<10<|a|<1 and put Za=Spec⁡O[U,V]/(UV−a)\mathcal Z_a=\operatorname{Spec}O[U,V]/(UV-a), with special-fibre origin oo. Suppose that there is a diagram (X,q) ⟵ (U,u) ⟶ (Za,o)(\mathcal X,q)\ \longleftarrow\ (\mathcal U,u) \ \longrightarrow\ (\mathcal Z_a,o) of flat finitely presented OO-schemes, with both arrows étale and all selected special-fibre points kk-rational. Then ]q[X^]q[_{\widehat{\mathcal X}} is an open annulus with coordinate TT satisfying ∣a∣<∣T∣<1|a|<|T|<1.

Proof. Apply Proposition 4.17.1 to both arrows. The formal completion of Za\mathcal Z_a is the standard nodal chart of Proposition 3.26.1. Its origin tube is computed in Proposition 3.20.1: with T=UT=U and V=a/TV=a/T, the two inequalities ∣U∣<1|U|<1 and ∣V∣<1|V|<1 become ∣a∣<∣T∣<1|a|<|T|<1. Transport this already established analytic identification along the two tube isomorphisms. ◻

4.18 Simultaneous splitting of finitely many nodes

Let KK be complete with its specified nontrivial non-Archimedean absolute value, and put O=K∘O=K^\circ and k=K~k=\widetilde K. Every finite extension below carries the unique absolute value extending this one.

Lemma 4.18.1. NAG-002052 (Lifting a finite Galois residue extension). For every finite Galois extension l/kl/k, there is a finite Galois extension L/KL/K and a specified kk-isomorphism L~≃l\widetilde L\simeq l such that [L:K]=[l:k],∣L×∣=∣K×∣,Gal⁡(L/K)→ ∼ Gal⁡(l/k).\begin{gathered} [L:K]=[l:k],\qquad |L^\times|=|K^\times|,\\ \operatorname{Gal}(L/K)\xrightarrow{\ \sim\ } \operatorname{Gal}(l/k). \end{gathered} The last map is induced by reduction.

Proof. Use the construction in Lemma 4.15.2. Write l=k(β)l=k(\beta), let p∈k[T]p\in k[T] be its monic minimal polynomial of degree dd, and lift pp to a monic P∈O[T]P\in O[T]. That proof constructs L=K(α)L=K(\alpha), with P(α)=0P(\alpha)=0, [L:K]=d[L:K]=d, L~=l\widetilde L=l, and α‾=β\overline\alpha=\beta. It also proves that PP is separable and that the value groups are equal. We retain these identifications.

Normality and separability of l/kl/k give the dd distinct roots β1,…,βd\beta_1,\ldots,\beta_d of pp in ll. Apply the simple-root construction of Lemma 4.15.1 in the complete field LL to each βj\beta_j. It gives a root αj∈L∘\alpha_j\in L^\circ of PP with reduction βj\beta_j. Distinct reductions make these roots distinct. Thus PP splits in LL, and L=K(α)L=K(\alpha) is its splitting field. By [StacksAlg] (Tag 09HU), L/KL/K is normal; it was already separable, so it is Galois.

Every KK-automorphism of LL preserves its absolute value by uniqueness in Lemma 1.12.1. It therefore induces a kk-automorphism of ll. Each α↦αj\alpha\mapsto\alpha_j defines a KK-embedding of LL into itself. Its image has degree dd over KK, so it is an automorphism. On the residue generator it induces β↦βj\beta\mapsto\beta_j. These are precisely the dd automorphisms of l/kl/k, and a KK-automorphism of LL is determined by the image of α\alpha. The dd constructed roots exhaust all roots of PP and have distinct reductions; thus an automorphism inducing the identity on ll fixes α\alpha. Reduction is therefore a bijective homomorphism, as claimed. ◻

Theorem 4.18.2. NAG-002053 (All lifts of a finite set of nodes). Let X→Spec⁡O\mathcal X\to\operatorname{Spec}O be finitely presented and at-worst-nodal of relative dimension one, and let q1,…,qsq_1,\ldots,q_s be a finite set of nodes of Xk\mathcal X_k. There is a finite Galois extension L/KL/K such that every point q′q' of XL~\mathcal X_{\widetilde L} over one of the qiq_i is a L~\widetilde L-rational split node. Writing O′=L∘O'=L^\circ, each such point admits an affine O′O'-scheme UU, a L~\widetilde L-rational point uu, and pointed étale maps (XO′,q′) ⟵ (U,u) ⟶ (Spec⁡O′[S,T]/(ST−a), 0‾),a∈L∘∘.\begin{gathered} (\mathcal X_{O'},q')\ \longleftarrow\ (U,u) \ \longrightarrow\ \bigl(\operatorname{Spec}O'[S,T]/(ST-a),\,\overline 0\bigr),\\ a\in L^{\circ\circ}. \end{gathered} Here 0‾\overline 0 is the special-fibre origin, whose prime ideal is (L∘∘,S,T)(L^{\circ\circ},S,T). If XK\mathcal X_K is smooth, every resulting parameter satisfies 0<∣a∣<10<|a|<1. One may take ∣L×∣=∣K×∣|L^\times|=|K^\times| and Gal⁡(L/K)≃Gal⁡(L~/k)\operatorname{Gal}(L/K)\simeq \operatorname{Gal}(\widetilde L/k) by reduction.

Proof. If the node set is empty, take L=KL=K. Otherwise, for each qiq_i, choose the original pointed normal-form diagram and point uiu_i used at the beginning of the proof of Theorem 4.15.3. Put ki=κ(qi)k_i=\kappa(q_i) and li=κ(ui)l_i=\kappa(u_i). That proof shows that ki/kk_i/k and li/kil_i/k_i are finite separable. Embed these fields compatibly into one separable closure of kk. Choose primitive generators for the finitely many li/kl_i/k by [StacksAlg] (Tag 030N), and let l/kl/k be the splitting field of their minimal polynomials. It is finite and normal by [StacksAlg] (Tag 09HU). Its generators are separable, so [StacksAlg] (Tag 09HC) makes it separable as well. Thus l/kl/k is finite Galois and contains every lil_i and its normal closure.

Apply Lemma 4.18.1 to obtain L/KL/K with L~=l\widetilde L=l. The points over qiq_i are the points of Spec⁡(ki⊗kl)\operatorname{Spec}(k_i\otimes_k l), and ki⊗kl→ ∼ ∏σ∈Hom⁡k(ki,l)l,b⊗c⟼(σ(b)c)σ.k_i\otimes_k l\xrightarrow{\ \sim\ } \prod_{\sigma\in\operatorname{Hom}_k(k_i,l)}l, \qquad b\otimes c\longmapsto(\sigma(b)c)_\sigma. Indeed, a primitive-element presentation writes the left side as l[T]/pi(T)l[T]/p_i(T), where pip_i splits into distinct linear factors. Evaluation at its roots is invertible by the Lagrange interpolation polynomials. Thus every point over qiq_i is rational and is indexed by an embedding σ:ki↪l\sigma:k_i\hookrightarrow l.

Fix such an embedding. By [StacksAlg] (Tag 0BME, part (2)), it extends to a kk-automorphism of ll, since l/kl/k is normal. Restriction to the chosen subfield lil_i gives an embedding τ:li↪l\tau:l_i\hookrightarrow l extending σ\sigma. The map Spec⁡l→Ui\operatorname{Spec}l\to U_i defined by τ\tau gives a rational point of (Ui)O′(U_i)_{O'} lying over the selected point q′q' indexed by σ\sigma. Its image in the étale base neighbourhood (Vi)O′(V_i)_{O'} is rational as well.

Now reuse the henselian-section step of the proof of Theorem 4.15.3, with this rational point and the fixed field LL. Lemma 4.15.1 supplies a section of (Vi)O′→Spec⁡O′(V_i)_{O'}\to\operatorname{Spec}O' through the point. Its image is open. Pulling the original diagram back along that section and then taking an affine neighbourhood of the rational point gives exactly the two pointed étale maps displayed in the statement. The remaining local arguments of that theorem apply unchanged: they identify the image in the standard chart with its origin, give a∈L∘∘a\in L^{\circ\circ}, identify the completed special local ring at q′q' with l[[S,T]]/(ST)l[[S,T]]/(ST), and prove a≠0a\ne 0 when the generic fibre is smooth. No enlargement of LL occurs in any of those steps.

There are only finitely many ii and finitely many embeddings for each ii. Applying the preceding construction to each one proves the assertion for every lift of every specified node, all over the same LL. The value-group and Galois assertions were already supplied by the first lemma. ◻

4.19 Coordinate independence on open annuli

Corollary 4.19.1. NAG-002032 (Open annuli and their radial homotopies). Let KK be a complete non-Archimedean field with its specified absolute value. Let X=AK(r,R),X′=AK(r′,R′),0≤r<R≤∞,0≤r′<R′≤∞\begin{gathered} X=A_K(r,R),\qquad X'=A_K(r',R'),\\ 0\leq r<R\leq\infty,\qquad 0\leq r'<R'\leq\infty \end{gathered} be standard open annuli with coordinates T,ST,S. Their radial strong homotopies are obtained by restricting and gluing the closed-annulus homotopies of Proposition 4.3.1. For any analytic isomorphism ϕ:X→X′\phi:X\to X', there are c∈K×c\in K^\times and ε∈{1,−1}\varepsilon\in\{1,-1\} such that ∣S(ϕ(x))∣=∣c∣ ∣T(x)∣ε(x∈X),ϕ(ζ0,ρ)=ζ0,∣c∣ρε′(r<ρ<R),ϕ(HX(x,u))=HX′(ϕ(x),u)(x∈X, 0≤u≤1).\begin{aligned} |S(\phi(x))|&=|c|\,|T(x)|^\varepsilon&& (x\in X),\\ \phi(\zeta_{0,\rho})&=\zeta'_{0,|c|\rho^\varepsilon} &&(r<\rho<R),\\ \phi\bigl(H_X(x,u)\bigr)&=H_{X'}\bigl(\phi(x),u\bigr) &&(x\in X,\ 0\leq u\leq1). \end{aligned} In particular, ϕ\phi preserves the radial retractions and induces a skeleton isometry t′=εt+v(c)t'=\varepsilon t+v(c) for t=−log⁡∣T∣t=-\log|T| and v(c)=−log⁡∣c∣v(c)=-\log|c|. When both boundary radii are positive and finite, the logarithmic modulus log⁡(R/r)\log(R/r) is invariant under analytic isomorphism.

Proof. Every point of XX belongs to the ordinary interior of some closed subannulus AK[a,b]A_K[a,b], with r<a<b<Rr<a<b<R. In the formula of Proposition 4.3.1, the homotopy keeps ∣T∣|T| fixed at every parameter value. Its seminorm formula at a point depends only on that point and uu, not on the chosen end radii. Consequently the homotopies on closed subannuli agree on overlaps. Their restrictions to the ordinary interiors give a continuous homotopy on X×[0,1]X\times[0,1]. The initial map, final radial retraction, and fixed-point identity on the Gauss skeleton hold on each subannulus, hence on XX. The same construction applies to X′X'.

Apply Corollary 4.10.3 to ϕ∗S\phi^*S and (ϕ−1)∗T(\phi^{-1})^*T. It gives constants c,d∈K×c,d\in K^\times and exponents m,n∈Zm,n\in\mathbb Z with ∣S(ϕ(x))∣=∣c∣ ∣T(x)∣m,ρ=∣d∣ ∣c∣nρmn(r<ρ<R).|S(\phi(x))|=|c|\,|T(x)|^m,\qquad \rho=|d|\,|c|^n\rho^{mn}\quad(r<\rho<R). The second identity may be evaluated at Gauss points of any two distinct radii. Thus mn=1mn=1, so m=n=ε∈{1,−1}m=n=\varepsilon\in\{1,-1\}. Surjectivity and the existence of all Gauss radii show that ρ↦∣c∣ρε\rho\mapsto|c|\rho^\varepsilon maps (r,R)(r,R) onto (r′,R′)(r',R').

Fix r<a<b<Rr<a<b<R and let [a′,b′][a',b'] be the increasing interval whose endpoints are ∣c∣aε|c|a^\varepsilon and ∣c∣bε|c|b^\varepsilon. The norm identities for ϕ\phi and its inverse identify the inverse image of AK[a′,b′]⊂X′A_K[a',b']\subset X' with AK[a,b]A_K[a,b] as a set. The universal property of analytic domains, proved in Theorem 2.40.8, restricts both morphisms and their inverse identities to these domains. Hence ϕ∣AK[a,b]:AK[a,b]→∼AK[a′,b′].\phi|_{A_K[a,b]}:A_K[a,b]\xrightarrow{\sim}A_K[a',b']. Theorem 4.16.1 applies to this nondegenerate closed-annulus isomorphism. Its Gauss-point and homotopy identities are the restrictions of the asserted ones: both homotopies preserve their radii and remain in the respective closed subannuli. These subannuli cover XX, proving all the identities. Negative logarithms give the skeleton isometry; its slope has absolute value one, so finite logarithmic interval lengths are preserved. ◻

4.20 The ends of residual tubes

Throughout this section KK is complete and nontrivially valued, O=K∘O=K^\circ and k=K~k=\widetilde K. A curve model is a flat finitely presented OO-scheme whose special fibre is a reduced pure one-dimensional scheme. Its formal generic fibre is assumed Hausdorff. We write vηv_\eta for the unique analytic point above a component generic point η\eta, as in Proposition 4.12.2.

Lemma 4.20.1. NAG-002054 (Boundary points of a closed-point tube). Let qq be a closed point of the special fibre of a curve model X\mathcal X, and put Dq=sp⁡−1(q)D_q=\operatorname{sp}^{-1}(q) in X=X^ηX=\widehat{\mathcal X}_\eta. Then DqD_q is open in XX, its closure CqC_q is compact, and Cq∖Dq ⊆ {vη:q∈{η}‾}.C_q\setminus D_q\ \subseteq\ \{v_\eta:q\in\overline{\{\eta\}}\}.

Proof. Choose an affine open U\mathcal U containing qq. Formal-open compatibility in Proposition 3.17.2 puts the entire tube DqD_q inside the compact affinoid U=U^ηU=\widehat{\mathcal U}_\eta. Since XX is Hausdorff, UU is closed in XX, so CqC_q is a closed subset of UU and is compact. On every formal affine chart, the inverse image of the closed residual point qq, when it is present, is given by finitely many inequalities ∣fi∣<1|f_i|<1; use Proposition 3.20.1 and the Noetherian special fibre. These inverse images are relatively open. The affinoid net topology of Proposition 3.17.2 therefore makes DqD_q open in XX.

For x∈Cqx\in C_q, write p=sp⁡(x)p=\operatorname{sp}(x) as a prime of the affine special fibre. If p⊈mqp\not\subseteq\mathfrak m_q, choose fˉ∈p∖mq\bar f\in p\setminus\mathfrak m_q and an integral lift ff. Then ∣f(x)∣<1|f(x)|<1, whereas ∣f(y)∣=1|f(y)|=1 for every y∈Dqy\in D_q. The relative open set ∣f∣<1|f|<1 separates xx from DqD_q, a contradiction. Thus p⊆mqp\subseteq\mathfrak m_q. In a pure one-dimensional Noetherian scheme the proper generalizations of a closed point are component generic points. Their analytic fibres are the singletons {vη}\{v_\eta\} by Proposition 4.12.2. This proves the inclusion. ◻

Lemma 4.20.2. NAG-002055 (Two components in a pointed nodal chart). Let Uk→Spec⁡k[s,t]/(st)U_k\to\operatorname{Spec}k[s,t]/(st) be an étale morphism and let u∈Uk(k)u\in U_k(k) map to the origin. In R=OUk,uR=\mathcal O_{U_k,u}, the ideals (s)(s) and (t)(t) are the two distinct minimal prime ideals. They identify the two distinct irreducible components of UkU_k through uu. Under any étale map from UkU_k to a reduced curve, their generic points map to component generic points; these two images may coincide.

Proof. The local map from the origin is flat, induces the identity on kk, and extends the maximal ideal to the maximal ideal. Thus [StacksAlg] (Tag 0AGX) identifies R^≃k[[s,t]]/(st).\widehat R\simeq k[[s,t]]/(st). The special-fibre ring RR is Noetherian. By [StacksAlg] (Tag 00MC), completion is faithfully flat, so R/(s)R/(s) and R/(t)R/(t) inject into k[[t]]k[[t]] and k[[s]]k[[s]], respectively. Consequently (s)(s) and (t)(t) are prime. Their extensions to R^\widehat R are incomparable, so the original ideals are incomparable. Their intersection is zero because it becomes zero in the displayed completion and R→R^R\to\widehat R is injective. Since st=0st=0, every prime contains (s)(s) or (t)(t). These are therefore exactly the minimal primes. The prime correspondence for localization identifies them with distinct global components through uu. Finally, an étale map is flat. Going down for a flat map, [StacksAlg] (Tag 00HS), implies that the contraction of a minimal prime is minimal: any strictly smaller prime would lift below the chosen minimal prime. This proves the last assertion. ◻

Proposition 4.20.3. NAG-002056 (Attaching the two ends of a nodal tube). Let X\mathcal X be a curve model and suppose given a pointed étale diagram (X,q)⟵(U,u)⟶(Wa,0ˉ),Wa=Spec⁡O[s,t]/(st−a),0<∣a∣<1,(\mathcal X,q)\longleftarrow(U,u)\longrightarrow (W_a,\bar0),\qquad W_a=\operatorname{Spec}O[s,t]/(st-a),\quad 0<|a|<1, where UU is affine and the three selected special-fibre points are kk-rational. In Y=U^ηY=\widehat U_\eta, put D=]u[D=]u[ and C=D‾C=\overline D. Let vt,vsv_t,v_s be the component points of the branches t=0,s=0t=0,s=0, respectively. Then D≃AK(∣a∣,1),C=D∪{vt,vs},∣s(vt)∣=1,∣s(vs)∣=∣a∣.D\simeq A_K(|a|,1),\qquad C=D\cup\{v_t,v_s\},\qquad |s(v_t)|=1,\quad |s(v_s)|=|a|. Every net in DD whose ∣s∣|s| tends to 11 converges to vtv_t; every net whose ∣s∣|s| tends to ∣a∣|a| converges to vsv_s. For each neighborhood of either endpoint in CC, a full radial tail of DD is contained in that neighborhood. The image of CC in XX is Cq=]q[‾C_q=\overline{]q[}; the only possible identification is that of vtv_t with vsv_s. The two endpoints map to the component points of the corresponding branches of qq.

Proof. The entire-tube isomorphisms of Corollary 4.17.3 identify ss with the annular coordinate. Write Us=Spec⁡SU_s=\operatorname{Spec}S and A=k[s,t]/(st)A=k[s,t]/(st). The given map U→WaU\to W_a induces the étale morphism Spec⁡S→Spec⁡A\operatorname{Spec}S\to\operatorname{Spec}A by base change, by [StacksAlg] (Tag 02GO). In particular SS is of finite type over kk. The two branch fields are Ft=k(s)F_t=k(s) and Fs=k(t)F_s=k(t). The map A→Ft×FsA\to F_t\times F_s is injective since (s)∩(t)=0(s)\cap(t)=0 in AA. Tensoring with SS gives the injective map S⟶(S⊗AFt)×(S⊗AFs).S\longrightarrow (S\otimes_A F_t)\times (S\otimes_A F_s). Indeed, for each prime q\mathfrak q of SS over p\mathfrak p of AA, the corresponding local map Ap→SqA_{\mathfrak p}\to S_{\mathfrak q} is flat by [StacksAlg] (Tag 02GU(1) and (2)). Localizing the displayed map at q\mathfrak q thus gives an injection at every prime. By [StacksAlg] (Tag 02GL(2)), each of the two affine fibres is a finite disjoint union of spectra of finite separable extensions of its branch field. Either fibre may be empty. This proves that SS is reduced.

If q\mathfrak q is minimal in SS, its contraction is minimal in AA. Otherwise the contraction contains an element outside both minimal primes of AA, by avoidance of these two primes. Such an element is a nonzerodivisor on SS by the displayed injection. But its image in the nonzero field SqS_{\mathfrak q} would be zero, a contradiction. For a minimal q\mathfrak q over the branch prime p\mathfrak p, localize S/qS/\mathfrak q at the nonzero elements of A/pA/\mathfrak p. The result is a nonzero domain quotient of the corresponding affine fibre algebra, hence a finite extension of that branch field. It is the fraction field of S/qS/\mathfrak q. Its transcendence degree over kk is one, so [StacksAlg] (Tag 00P0) gives dim⁡(S/q)=1\dim(S/\mathfrak q)=1. Thus the special fibre of UU is a reduced pure one-dimensional curve. Lemma 4.20.2 gives exactly the two components through uu. Lemma 4.20.1 gives C∖D⊆{vt,vs}C\setminus D\subseteq\{v_t,v_s\}. At vtv_t, the reduction of ss is nonzero, so ∣s(vt)∣=1|s(v_t)|=1; at vsv_s, the reduction of tt is nonzero, so ∣t(vs)∣=1|t(v_s)|=1 and st=ast=a gives ∣s(vs)∣=∣a∣|s(v_s)|=|a|.

The affinoid YY is compact Hausdorff. A net in DD whose ∣s∣|s| tends to 11 has cluster points in CC, and continuity of ∣s∣|s| forces every cluster point to be vtv_t. Compactness then implies convergence to vtv_t: otherwise a subnet outside a fixed neighborhood would have a cluster point there. The inner end is identical. Applying this to the Gauss points with radii tending to the two endpoints proves that both vtv_t and vsv_s belong to CC. If NN is a neighborhood of vtv_t in CC, the compact set C∖NC\setminus N has max⁡∣s∣<1\max|s|<1 after replacing NN by an open subneighborhood. Thus a full outer tail lies in NN. The minimum of ∣s∣|s| on a compact complement similarly proves the inner-tail assertion.

By Lemma 4.20.2, the two residual generic points map to component generic points of Xs\mathcal X_s. Functoriality of specialization and uniqueness of component points identify the images of vt,vsv_t,v_s. The image of compact CC is closed in Hausdorff XX and contains ]q[]q[. Conversely, continuity and density of DD in CC put that image inside ]q[‾\overline{]q[}. The restriction D→]q[D\to]q[ is an isomorphism, and an endpoint cannot map into ]q[]q[ because its specialization is generic. Consequently only the two endpoints can be identified. ◻

Corollary 4.20.4. NAG-002057 (The complete nodal homotopy). Under the hypotheses of Proposition 4.20.3, the radial homotopy on ]q[]q[ extends uniquely by fixing the attached component points to a jointly continuous strong deformation retraction of CqC_q onto a closed edge of length −log⁡∣a∣-\log|a|. If the endpoint images coincide, this edge is a circle with two distinct incident half-edges at its vertex. The construction is independent of the chosen pointed nodal chart and annular coordinate.

Proof. First work in C=D∪{vt,vs}C=D\cup\{v_t,v_s\} of the proposition. The annulus homotopy of Proposition 4.3.1 preserves ∣s∣|s|. The full-tail assertion shows that fixing vt,vsv_t,v_s extends this homotopy continuously at both endpoints, uniformly in its parameter. The resulting map C×[0,1]→CC\times[0,1]\to C is therefore jointly continuous. The Gauss radius map extends to a continuous injection of [∣a∣,1][|a|,1] into CC, with endpoints vs,vtv_s,v_t; compactness identifies its image with a closed interval. The extended homotopy fixes this interval and retracts onto it.

The map C→CqC\to C_q is a compact-to-Hausdorff surjection, as is its product with [0,1][0,1], so both are quotient maps. The only possible identified pair consists of the two fixed endpoints. The homotopy therefore descends to a continuous map on Cq×[0,1]C_q\times[0,1], with all strong-retraction identities unchanged. The quotient of the Gauss interval is an interval or, when its endpoints coincide, a circle. Its logarithmic length is −log⁡∣a∣-\log|a| by Corollary 4.16.2. The full homotopy commutes with every open-annulus coordinate change by Theorem 4.19.1; continuity on the dense tube gives the same equality at the ends. This proves independence and retains both half-edges of a loop. ◻

4.21 Canonical contraction of a smooth tube

Lemma 4.21.1. NAG-002058 (The compactification of a rational smooth tube). Let q∈Xs(k)q\in\mathcal X_s(k) be a point where a curve model X\mathcal X is smooth of relative dimension one. If η\eta is the unique component generic point specializing to qq, then Dq=]q[≃{x:∣T(x)∣<1},Cq=Dq‾=Dq∪{vη}.D_q=]q[\simeq\{x:|T(x)|<1\},\qquad C_q=\overline{D_q}=D_q\cup\{v_\eta\}. This closure is the one-point compactification of DqD_q. In each open-unit-disc coordinate TT, the sets {vη}∪{x∈Dq:∣T(x)∣>r},0<r<1,\{v_\eta\}\cup\{x\in D_q:|T(x)|>r\},\qquad 0<r<1, form a neighborhood basis of vηv_\eta in CqC_q.

Proof. Corollary 4.17.2 gives the disc isomorphism. Smoothness makes the local ring of the special-fibre curve at qq a regular one-dimensional local ring, hence a domain; equivalently its completed pointed étale coordinate ring is k[[T]]k[[T]] and faithful flatness of completion gives a unique minimal prime. Thus exactly one component passes through qq. Lemma 4.20.1 shows that the compact closure can add only vηv_\eta. The open unit disc is not compact: the continuous function ∣T∣|T| has supremum 11 and does not attain it, as witnessed by its Gauss points of radii r<1r<1. Therefore the closure must add vηv_\eta. A compact Hausdorff space obtained by adjoining one point to an open locally compact space is its one-point compactification: the complements of neighborhoods of the added point are compact subsets of the open space, and conversely every such compact subset is closed in the compactification.

For r<1r<1, the closed subdisc ∣T∣≤r|T|\leq r is compact, so its complement in CqC_q is a neighborhood of vηv_\eta. Conversely, every compact subset of DqD_q has a maximum ∣T∣<1|T|<1. These subdiscs are therefore cofinal among compact subsets, proving the asserted neighborhood basis. ◻

Theorem 4.21.2. NAG-002059 (The intrinsic open-disc homotopy). Let DD be the open unit disc over a complete nontrivially valued field KK, and let D∗=D∪{v}D^*=D\cup\{v\} be its one-point compactification. Write P(T+Z)=∑j≥0Pj(T)Zj(P∈K[T]).P(T+Z)=\sum_{j\geq0}P_j(T)Z^j\qquad(P\in K[T]). There is a jointly continuous strong deformation retraction HD:D∗×[0,1]→D∗H_D:D^*\times[0,1]\to D^* onto {v}\{v\}, characterized for x∈Dx\in D and 0≤u<10\leq u<1 by ∣P(HD(x,u))∣=max⁡j∣Pj(x)∣uj,HD(x,1)=v,HD(v,u)=v.|P(H_D(x,u))|=\max_j |P_j(x)|u^j, \qquad H_D(x,1)=v,\qquad H_D(v,u)=v. Every analytic isomorphism of open unit discs, including an isomorphism semilinear for an isometric field automorphism, commutes with this entire parametrized homotopy. The homotopy also commutes with projection under any complete isometric ground-field extension. In particular it gives a coordinate-independent contraction of CqC_q in Lemma 4.21.1.

Proof. For u<1u<1, evaluate at T(x)+ZT(x)+Z over H(x)\mathcal H(x) and then take the radius-uu Gauss seminorm. It is multiplicative and has ∣T(HD(x,u))∣=max⁡{∣T(x)∣,u}<1.|T(H_D(x,u))|=\max\{|T(x)|,u\}<1. Thus it defines a point of DD, and at u=0u=0 it gives xx. On a neighborhood of any (x,u)∈D×[0,1)(x,u)\in D\times[0,1), choose r<1r<1 strictly larger than both ∣T(x)∣|T(x)| and uu. The displayed formula is continuous on polynomials and remains bounded on the closed radius-rr disc; uniform polynomial approximation proves continuity for its affinoid algebra. This proves joint continuity on D×[0,1)D\times[0,1). At vv, the preceding inequality preserves every tail ∣T∣>r|T|>r, uniformly in uu. As u→1u\to 1, it also puts all images in ∣T∣>r|T|>r for u>ru>r, uniformly in xx. The neighborhood basis of Lemma 4.21.1, whose one-point-compactification argument applies equally to D∗D^*, proves joint continuity at every remaining point. The strong contraction identities follow from the construction.

We check coordinate independence explicitly. An analytic isomorphism is represented by a power series g(T)=∑n≥0anTng(T)=\sum_{n\geq 0}a_nT^n convergent on every closed disc of radius less than one. Such an expansion follows by restricting to those discs and using uniqueness of Tate coefficients on overlaps. Since ∣g∣<1|g|<1 on DD, evaluation at its Gauss points and passage of their radii to 11 give ∣an∣≤1|a_n|\leq 1 for every nn and ∣a0∣<1|a_0|<1. The inverse has the same coefficient bounds. The derivative identity for the two inverse series at 00 implies ∣a1∣=1|a_1|=1: both ∣g′(0)∣|g'(0)| and the derivative of the inverse at a0a_0 are at most one and their product is one.

Put F=H(x)F=\mathcal H(x), α=T(x)\alpha=T(x) and β=g(x)\beta=g(x). For 0<u<10<u<1, expansion over FF gives g(α+Z)=β+∑j≥1bjZj,∣b1∣=1,∣bj∣≤1 (j≥2).g(\alpha+Z)=\beta+\sum_{j\geq1}b_jZ^j, \qquad |b_1|=1,\quad |b_j|\leq1\ (j\geq2). Indeed, bj=∑n≥jan(nj)αn−jb_j=\sum_{n\geq j}a_n\binom nj\alpha^{n-j}; these sums converge, and for j=1j=1 all terms after a1a_1 have norm at most ∣α∣<1|\alpha|<1. On the radius-uu disc, g(α+Z)−β=b1Z(1+E(Z)),∥E∥u≤u<1.g(\alpha+Z)-\beta=b_1Z(1+E(Z)),\qquad \|E\|_u\leq u<1. For a polynomial Q(S)Q(S) write Q(β+W)=∑jqjWjQ(\beta+W)=\sum_jq_jW^j and put M=max⁡j∣qj∣ujM=\max_j|q_j|u^j. Replacing W=b1Z(1+E)W=b_1Z(1+E) by b1Zb_1Z changes its Gauss norm by an error at most uM<MuM<M when Q≠0Q\ne 0. The Gauss norm of ∑jqjb1jZj\sum_jq_jb_1^jZ^j is MM. Thus ∥Q(g(α+Z))∥u=max⁡j∣qj∣uj.\bigl\|Q(g(\alpha+Z))\bigr\|_u =\max_j|q_j|u^j. These are precisely the polynomial evaluations at g(HD(x,u))g(H_D(x,u)) and HD(g(x),u)H_D(g(x),u). Points of the open disc are determined by polynomial evaluations, so the two points agree. The identities at u=0,1u=0,1 follow from the endpoints and the extension of a disc isomorphism to its one-point compactification. An isometric automorphism of the coefficient field preserves all these formulas. Finally, under any complete isometric extension, the formulas agree on the original polynomials by the isometric embedding of completed residue fields. At the added endpoint the projection sends the endpoint to the endpoint; continuity follows also from its preservation of the radial tails. This proves all the claimed compatibilities. ◻

4.22 Smooth tubes with a separable residue field

Lemma 4.22.1. NAG-002060 (The integral algebra of a residue-field lift). Let l/kl/k be finite separable and let L=K(α)L=K(\alpha) be the lift constructed in Lemma 4.15.2: the minimal polynomial P∈O[T]P\in O[T] of α\alpha is monic and reduces to the separable minimal polynomial of a generator of l/kl/k. Put d=[l:k]d=[l:k]. Then ∣∑i=0d−1ciαi∣=max⁡i∣ci∣,L∘=⨁i=0d−1Oαi=O[T]/(P),L∘∘=K∘∘L∘.\left|\sum_{i=0}^{d-1}c_i\alpha^i\right|=\max_i|c_i|, \qquad L^\circ=\bigoplus_{i=0}^{d-1}O\alpha^i=O[T]/(P), \qquad L^{\circ\circ}=K^{\circ\circ}L^\circ. For every flat finitely presented OO-scheme X\mathcal X, its formal completion and analytic generic fibre commute with this coefficient extension. On a completed affine algebra A0A_0 the corresponding formulas are B0=A0⊗OL∘=⨁i=0d−1A0αi,B0[1/π]=A⊗^KL,B0/L∘∘B0=(A0/K∘∘A0)⊗kl.\begin{aligned} B_0&=A_0\otimes_O L^\circ=\bigoplus_{i=0}^{d-1}A_0\alpha^i,\\ B_0[1/\pi]&=A\widehat\otimes_K L,\\ B_0/L^{\circ\circ}B_0&=(A_0/K^{\circ\circ}A_0)\otimes_k l. \end{aligned} If the original special fibre is reduced, so is the extended special fibre. Specialization commutes with the projection, and the inverse image of a closed-point tube is the disjoint union of the tubes of the points above that closed point.

Proof. For a nonzero coefficient vector, divide by a coefficient of largest absolute value. All normalized coefficients lie in OO, and at least one has nonzero reduction. Linear independence of 1,αˉ,…,αˉd−11,\bar\alpha,\ldots,\bar\alpha^{d-1} over kk makes the reduction of this normalized sum nonzero, hence its absolute value is one. This proves the norm formula. It proves the two integral identifications as well, since membership in the closed or open unit ball is equivalent to that property for every coefficient.

A finite free direct sum commutes with every quotient by πn\pi^n and with the inverse limit defining completion. Thus for an affine algebraic chart, completing its base change gives exactly B0B_0; Lemma 4.13.1 makes this an admissible model. After inverting π\pi, the finite coordinate comparison in the proof of Lemma 4.9.1, which uses only finiteness of L/KL/K, identifies the algebraic tensor with the completed tensor. The displayed special-fibre formula follows from L∘∘=K∘∘L∘L^{\circ\circ}=K^{\circ\circ}L^\circ. These constructions respect affine restriction, so they glue.

For completeness, separability preserves the required reducedness directly. A reduced finite-type kk-algebra RR injects into the finite product of the fraction fields of its minimal-prime quotients. Tensoring with ll, which is flat over kk, preserves this injection. Each field tensor with l=k[T]/(p)l=k[T]/(p) is reduced: pp remains relatively prime to its derivative and hence factors into distinct irreducible factors. Thus R⊗klR\otimes_k l embeds in a product of reduced rings and is reduced. Finally, specialization is defined by the elements of absolute value less than one. The displayed algebra maps therefore make it commute with the projection. Taking inverse images of a singleton residual point gives the asserted tube decomposition. ◻

Proposition 4.22.2. NAG-002061 (Descent of the smooth-tube contraction). Let qq be a closed point where a curve model X/O\mathcal X/O is smooth of relative dimension one, and suppose κ(q)/k\kappa(q)/k is finite separable. Let η\eta be its component generic point. Then ]q[‾=]q[∪{vη}.\overline{]q[}=]q[\cup\{v_\eta\}. This compact space admits a canonical jointly continuous strong deformation retraction onto vηv_\eta. After any finite Galois residue-field lift which splits qq, it is the descent of the canonical open-disc contractions on the finitely many rational tubes. It is independent of that lift and is functorial for isomorphisms of curve models and isometric automorphisms of the ground field.

Proof. Take a finite Galois extension l/kl/k containing the normal closure of κ(q)/k\kappa(q)/k, and use Lemma 4.18.1 to obtain L/KL/K. The separable polynomial factorization gives κ(q)⊗kl≃∏ι:κ(q)↪ll\kappa(q)\otimes_k l\simeq\prod_{\iota:\kappa(q)\hookrightarrow l}l. Consequently the points q1,…,qmq_1,\ldots,q_m over qq are rational smooth points of the extended model. The preceding lemma identifies the generic-fibre projection p:XL→Xp:X_L\to X and gives p−1(Dq)=∐j=1mDqj.p^{-1}(D_q)=\coprod_{j=1}^m D_{q_j}. Each DqjD_{q_j} has compact closure Cj=Dqj∪{vj}C_j=D_{q_j}\cup\{v_j\} by Lemma 4.21.1. By flat going down ([StacksAlg] (Tag 00HS)), each generic point above the unique component through qq maps to a component generic point; its image is necessarily η\eta. Thus p(vj)=vηp(v_j)=v_\eta. Continuity puts this point in Dq‾\overline{D_q}, and Lemma 4.20.1 gives Cq=Dq∪{vη}C_q=D_q\cup\{v_\eta\}.

The projection pp is open by Theorem 4.9.2. Any continuous open surjection satisfies p−1(E‾)=p−1(E)‾p^{-1}(\overline E)=\overline{p^{-1}(E)}: an open neighborhood upstairs maps to an open neighborhood downstairs, which meets EE whenever its centre lies in E‾\overline E. Apply this to E=DqE=D_q to obtain p−1(Cq)=⋃j=1mCj.p^{-1}(C_q)=\bigcup_{j=1}^m C_j. The canonical homotopies of Theorem 4.21.2 agree on intersections of these closed sets, which consist only of fixed component points. Finite closed pasting gives a continuous homotopy on p−1(Cq)p^{-1}(C_q). Its full coordinate and semilinear invariance makes it Gal⁡(L/K)\operatorname{Gal}(L/K)-equivariant. The compact-to-Hausdorff surjection p−1(Cq)→Cqp^{-1}(C_q)\to C_q remains a quotient map after product with [0,1][0,1]. Hence the homotopy descends, and its endpoint and fixed-point identities give the asserted strong contraction.

To compare two choices of lift, take a finite Galois extension of kk containing both residue fields and lift it in the same way. The simple roots of each of the two defining polynomials lift in this common complete field; sending the original generators to these roots gives compatible isometric embeddings of both lifted fields. The ground-field compatibility in Theorem 4.21.2 makes the two homotopies equal after this common extension. Surjectivity of projection makes their descents equal. The same argument together with coordinate and semilinear invariance proves functoriality. ◻

4.23 Component stars and the global skeleton

Lemma 4.23.1. NAG-002062 (Cofinitely many whole tubes in a vertex neighborhood). Let X\mathcal X be a curve model, let η\eta be a component generic point of Xs\mathcal X_s, and let v=vηv=v_\eta. For every ordinary open neighborhood NN of vv in X=X^ηX=\widehat{\mathcal X}_\eta, there are only finitely many closed points q∈{η}‾q\in\overline{\{\eta\}} for which the whole tube Dq=sp⁡−1(q)D_q=\operatorname{sp}^{-1}(q) is not contained in NN.

Proof. Take a finite affine cover of X\mathcal X. Every member meeting {η}‾\overline{\{\eta\}} contains η\eta, since an open set contains the generalizations of each of its points. Thus it suffices to prove the assertion inside a formal affine chart U=Spf⁡A0\mathfrak U=\operatorname{Spf}A_0 containing η\eta. Write A=A0[1/π]A=A_0[1/\pi] and R=A0/mA0R=A_0/\mathfrak m A_0; let P⊂RP\subset R be the minimal prime of η\eta.

Choose gˉ∈R∖P\bar g\in R\setminus P which belongs to every other minimal prime of RR. Such an element is obtained by multiplying, for each of the finitely many other minimal primes P′P', an element of P′∖PP'\setminus P; if there are no others, take gˉ=1\bar g=1. The reduced ring RgˉR_{\bar g} has the unique minimal prime PRgˉP R_{\bar g}, hence that prime is zero and RgˉR_{\bar g} is an integral domain. The complement of D(gˉ)D(\bar g) on the component of η\eta is finite: a nonzero regular function on a finite-type integral curve vanishes on a zero-dimensional Noetherian closed subset, which has finitely many points.

Lift gˉ\bar g to g∈A0g\in A_0 and use the principal formal open B0=A0{g−1}B_0=A_0\{g^{-1}\}. Proposition 3.17.1 identifies its generic fibre with ∣g∣=1|g|=1, and its special fibre is Spec⁡Rgˉ\operatorname{Spec}R_{\bar g}. Put B=B0[1/π]B=B_0[1/\pi]. By Lemma 4.12.1, the spectral norm on BB is attained by an integral presentation and its nonzero values belong to ∣K×∣|K^\times|. By Proposition 4.12.2, this affinoid has the singleton Shilov boundary {v}\{v\}. In particular, for f∈Af\in A, its restriction fBf_B satisfies ∥fB∥=∣f(v)∣.\|f_B\|=|f(v)|. If this value is zero, fB=0f_B=0 and the value of ff is zero throughout every tube over D(gˉ)D(\bar g). Otherwise choose c∈K×c\in K^\times with ∣c∣=∣f(v)∣|c|=|f(v)|. Then b=fB/cb=f_B/c belongs to B0B_0, and its reduction is a nonzero element of the integral curve ring RgˉR_{\bar g}. Outside its finitely many zeros, specialization gives ∣b(x)∣=1|b(x)|=1 for every point of the entire tube, and therefore ∣f(x)∣=∣f(v)∣(x∈Dq)|f(x)|=|f(v)|\qquad(x\in D_q) for all but finitely many closed qq of this component in the chart. Here the whole tube lies in the chosen principal formal open by formal-open compatibility.

The topology of M(A)\mathcal M(A) is generated by the evaluation functions x↦∣f(x)∣x\mapsto |f(x)|. Choose a basic neighborhood of vv inside N∩M(A)N\cap\mathcal M(A), using finitely many such functions. The preceding equality puts all but finitely many whole tubes in this basic neighborhood. Finally take the union of the finite exceptional sets over the chosen finite affine cover. ◻

Proposition 4.23.2. NAG-002063 (A compact component star and its contraction). Let X\mathcal X be a curve model, and let SηS_\eta be the set of closed points on the component of η\eta where X\mathcal X is smooth of relative dimension one. Assume that κ(q)/k\kappa(q)/k is separable for every q∈Sηq\in S_\eta. Then Aη={vη}∪⋃q∈SηDqA_\eta=\{v_\eta\}\cup\bigcup_{q\in S_\eta}D_q is a compact closed subspace of XX. The canonical contractions of the completed tubes Cq=Dq∪{vη}C_q=D_q\cup\{v_\eta\} paste to a jointly continuous strong deformation retraction Hη:Aη×[0,1]→AηH_\eta:A_\eta\times[0,1]\to A_\eta onto {vη}\{v_\eta\}.

Proof. Proposition 4.22.2 identifies the compact closure of every DqD_q with CqC_q and supplies its canonical contraction. Distinct CqC_q meet only at v=vηv=v_\eta, which their homotopies fix. Thus there is a well-defined pasted map HηH_\eta.

To prove compactness, take an open cover of AηA_\eta and a member containing vv. It is the intersection with an open neighborhood of vv in XX. By Lemma 4.23.1, it contains all but finitely many whole DqD_q. Each of the finitely many remaining CqC_q is compact. Finite subcovers of those CqC_q, together with the chosen member, cover AηA_\eta. This proves compactness; closedness follows from the Hausdorff property of XX.

Continuity away from vv follows because every DqD_q is open in XX. For continuity at vv, let NN be a relatively open neighborhood of vv in AηA_\eta. There is a finite subset E⊂SηE\subset S_\eta such that Cq⊂NC_q\subset N for q∉Eq\notin E. For each q∈Eq\in E, continuity on Cq×[0,1]C_q\times[0,1], together with compactness of [0,1][0,1] and the equality Hq(v,u)=vH_q(v,u)=v, gives a relatively open neighborhood MqM_q of vv in CqC_q such that Hq(Mq×[0,1])⊂N.H_q(M_q\times[0,1])\subset N. Indeed, finitely many product neighborhoods of the compact set {v}×[0,1]\{v\}\times[0,1] suffice, and the intersection of their first factors is such an MqM_q. The finite union F=⋃q∈E(Cq∖Mq)F=\bigcup_{q\in E}(C_q\setminus M_q) is compact, closed in AηA_\eta, and does not contain vv. Hence M=Aη∖FM=A_\eta\setminus F is a neighborhood of vv and Hη(M×[0,1])⊂N.H_\eta(M\times[0,1])\subset N. This proves joint continuity uniformly in the homotopy parameter, despite the possibly infinite set SηS_\eta. The strong deformation retraction identities hold on every CqC_q and at vv, so they hold globally. ◻

Theorem 4.23.3. NAG-002064 (The skeleton of a split curve model). Let X\mathcal X be a curve model. Assume that every closed point of its special fibre is either a point where X\mathcal X is smooth of relative dimension one with finite separable residue field over kk, or one of finitely many kk-rational nodes qq admitting a pointed diagram as in Proposition 4.20.3, with parameter aq∈Oa_q\in O, 0<∣aq∣<10<|a_q|<1. Then X=X^ηX=\widehat{\mathcal X}_\eta has a canonical finite metric graph Σ(X)\Sigma(\mathcal X) with vertices vηv_\eta and one edge of length −log⁡∣aq∣-\log|a_q| for every node qq. The ends of this edge attach to the component points of the two branches at qq, allowing a self-loop. There is a jointly continuous strong deformation retraction HX:X×[0,1]⟶X,HX(X,1)=Σ(X).H_{\mathcal X}:X\times[0,1]\longrightarrow X, \qquad H_{\mathcal X}(X,1)=\Sigma(\mathcal X). On each nodal tube it is the radial annulus homotopy; on each completed smooth tube it is the canonical disc contraction of Proposition 4.22.2. The entire parametrized homotopy is independent of the pointed charts and commutes with model isomorphisms, including those semilinear for isometric ground-field automorphisms.

Proof. The vertex set V={vη}V=\{v_\eta\} is finite by Proposition 4.12.2. For every node qq, let Iq=[0,−log⁡∣aq∣]I_q=[0,-\log|a_q|] and map its interior to the Gauss skeleton of DqD_q by the coordinate −log⁡∣s∣-\log|s|. Proposition 4.20.3 and Corollary 4.20.4 continuously extend this map to the two incident vertices, including when these coincide. Form a finite graph GG from the disjoint union of VV and the IqI_q by these endpoint identifications. Its map to XX is continuous and injective: distinct interiors specialize to distinct nodes, and no interior point specializes to a component generic point. Since GG is compact and XX is Hausdorff, this is a homeomorphism onto its image, denoted Σ(X)\Sigma(\mathcal X). Give each edge its interval length. Corollary 4.16.2 makes the length independent of the pointed coordinate, using the original absolute value on KK.

Every point of the special fibre is a component generic point or a closed point. Thus XX is the finite union of the component stars AηA_\eta and the completed nodal tubes CqC_q. Proposition 4.23.2 makes the former compact, and Lemma 4.20.1 makes the latter compact. They are therefore closed in XX. Their pairwise intersections consist only of incident vertices: the open tubes have distinct specializations, and their boundaries were identified in the endpoint propositions.

On each AηA_\eta use HηH_\eta, and on each CqC_q use Corollary 4.20.4. These homotopies fix all intersection vertices. The finite closed pasting lemma, applied to their products with [0,1][0,1], gives a jointly continuous map HXH_{\mathcal X}. At time zero it is the identity; at time one it maps each star to its vertex and each nodal tube to its Gauss edge. It fixes all these vertices and edges at every time. These are exactly the strong deformation retraction identities.

Coordinate independence on smooth tubes is part of Proposition 4.22.2; on nodal tubes it is Corollary 4.19.1. A model isomorphism permutes residual components and closed points, sends their completed tubes to the corresponding completed tubes, and induces the indicated analytic or semilinear isomorphisms. Hence it commutes with every local homotopy, and therefore with their global pasting. ◻

Corollary 4.23.4. NAG-002065 (Finite Galois descent of the model skeleton). Let L/KL/K be a finite Galois extension of complete nontrivially valued fields, using the extension of the fixed absolute value on KK, and let G=Gal⁡(L/K)G=\operatorname{Gal}(L/K). Suppose that a curve model Y\mathcal Y over L∘L^\circ satisfies Theorem 4.23.3, its formal generic fibre is identified with XLX_L, and the natural semilinear GG-action on XLX_L extends to Y\mathcal Y. Then the model homotopy descends to a jointly continuous strong deformation retraction of XX onto the embedded finite graph Σ(Y)/G\Sigma(\mathcal Y)/G.

Proof. The model homotopy is GG-equivariant by the final assertion of Theorem 4.23.3. Theorem 4.9.2 identifies XL/GX_L/G with XX and the quotient of its invariant skeleton with the skeleton’s image in XX. Corollary 4.9.3 then descends the entire homotopy and all strong retraction identities.

For completeness, this quotient is a finite graph. The GG-action permutes vertices and edges; on an edge it is an isometry by annulus-coordinate invariance, hence, after identifying its interval, is either orientation-preserving or a reflection. Subdivide every edge at its midpoint. An element stabilizing a resulting edge cannot reverse it, because original vertices and new midpoints are distinct GG-invariant classes. The quotient is therefore the finite graph whose vertices and edges are the corresponding orbits. Its edge lengths are the lengths of these subdivided intervals. Thus an inverted original edge has an image of half its length, whereas an edge with no inversion retains its total length after the unnecessary subdivision is removed. No field-degree rescaling is introduced. ◻

4.24 Residual functions and outgoing slopes

Let KK be complete and nontrivially valued. We retain its given absolute value and write O=K∘O=K^\circ, m=K∘∘\mathfrak m=K^{\circ\circ}, and k=O/mk=O/\mathfrak m.

Lemma 4.24.1. NAG-002069 (Normalization points and model branches). Let DD be a reduced finite-type curve over a field kk, with at-worst-nodal singularities. Its normalization is the disjoint union of the normalizations CνC^\nu of its irreducible components CC. A smooth closed point qq of DD has one preimage on the normalization, on the unique component through qq, with the same residue field. A split node qq has precisely two preimages p+,p−p_+,p_-, both with residue field kk. The two points lie on the normalized components selected by the two local branches; they remain distinct if both branches belong to the same global component.

For a pointed étale diagram with EE affine (D,q) ⟵ (E,u) ⟶ (Spec⁡k[s,t]/(st),o),κ(q)=κ(u)=k,(D,q)\ \longleftarrow\ (E,u)\ \longrightarrow\ \bigl(\operatorname{Spec}k[s,t]/(st),o\bigr), \qquad \kappa(q)=\kappa(u)=k, the normalization points over uu identify with those over qq and with the two origins of the normalized coordinate axes. Thus the branches t=0t=0 and s=0s=0 select unambiguously the two points of the normalization of DD above qq.

Proof. On a reduced affine chart, the total ring of fractions is the product of the fraction fields of the minimal-prime quotients. Its integral closure is the product of the integral closures in those fields; the product idempotents are integral. These descriptions localize and give the asserted component decomposition. Finiteness of normalization is [StacksAlg] (Tag 0BXR). At a smooth point the local ring is regular, and hence a normal discrete valuation ring by [StacksAlg] (Tag 0B8Y). The normalization is unchanged there. For a split node, the assertion about the two distinct rational normalization points is precisely [StacksAlg] (Tag 0CBW).

Normalization commutes with the two étale maps by [StacksAlg] (Tag 03GV). To apply that relative-normalization statement, use the quasi-compact and quasi-separated map from the finite disjoint union of component generic points to the reduced curve. Its étale base change is a finite disjoint union of generic-point fields, so the relative normalizations are the ordinary ones. The fibre of each Cartesian normalization square at the rational point uu is the two-point normalization fibre at its image. The normalization of k[s,t]/(st)k[s,t]/(st) is k[s]×k[t]k[s]\times k[t]: each factor is integrally closed, the product is finite over the original ring, and both rings have the same total ring of fractions. Its origins are the branches t=0t=0 and s=0s=0. This gives the specified identifications. ◻

Proposition 4.24.2. NAG-002070 (The valued field of a component point). Let A0A_0 be an admissible OO-algebra with reduced special fibre, let A=A0[1/π]A=A_0[1/\pi], and let x=xηx=x_\eta be the component point corresponding to a minimal prime P⊂A~P\subset\widetilde A. The reduction map induces an isomorphism Frac⁡(A~/P)→ ∼ H(x)~,∣H(x)×∣=∣K×∣.\operatorname{Frac}(\widetilde A/P) \xrightarrow{\ \sim\ }\widetilde{\mathcal H(x)}, \qquad |\mathcal H(x)^\times|=|K^\times|. If η\eta is the generic point of a residual curve component CC, the left hand field is k(C)k(C). In particular xx has residual transcendence degree one and relative rational value-group rank zero.

Proof. By Proposition 4.12.2, the Shilov boundary consists of finitely many component points x=x1,x2,…,xrx=x_1,x_2,\ldots,x_r, with distinct minimal reduction primes P=P1,P2,…,PrP=P_1,P_2,\ldots,P_r. For each j>1j>1, choose h‾j∈Pj∖P\overline h_j\in P_j\setminus P and lift their product to h∈A0h\in A_0. If r=1r=1, take h=1h=1. Then ∣h(x)∣=1,∣h(xj)∣<1(j>1).|h(x)|=1,\qquad |h(x_j)|<1\quad(j>1). For a∈Aa\in A with a(x)≠0a(x)\ne 0, choose NN so large that ∣a(xj)h(xj)N∣<∣a(x)∣|a(x_j)h(x_j)^N|<|a(x)| for every j>1j>1. The boundary maximum formula gives ρA(ahN)=∣a(x)∣\rho_A(ah^N)=|a(x)|. Lemma 4.12.1, including the norm-attaining lift used there, implies that this nonzero spectral value belongs to ∣K×∣|K^\times|. Choose c∈K×c\in K^\times of this absolute value. Then ahN/c∈A0ah^N/c\in A_0 and its reduction is outside PP.

The homomorphism A~/P→H(x)~\widetilde A/P\to \widetilde{\mathcal H(x)} is injective by the definition of PP, so it extends to its fraction field. To prove surjectivity, first consider a/ba/b with a,b∈Aa,b\in A, a(x)b(x)≠0a(x)b(x)\ne 0, and ∣a(x)/b(x)∣=1|a(x)/b(x)|=1. Choose a single NN large enough for both aa and bb in the preceding argument, and choose ca,cb∈K×c_a,c_b\in K^\times with ∣ca∣=∣a(x)∣|c_a|=|a(x)| and ∣cb∣=∣b(x)∣|c_b|=|b(x)|. Both a′=ahN/ca,b′=bhN/cba'=ah^N/c_a,\qquad b'=bh^N/c_b belong to A0A_0 and have nonzero residues on CC. Since ∣ca/cb∣=1|c_a/c_b|=1, the residue of a/b=(ca/cb)a′/b′a/b=(c_a/c_b)a'/b' lies in Frac⁡(A~/P)\operatorname{Frac}(\widetilde A/P). Every unit of H(x)\mathcal H(x) is within distance strictly less than one of such a fraction, because H(x)\mathcal H(x) is the completion of Frac⁡(A/ker⁡x)\operatorname{Frac}(A/\ker x). It therefore has the same residue as that fraction. This proves surjectivity.

The first paragraph gives the value-group assertion for fractions. A nonzero limit in a non-Archimedean valued field has the same absolute value as every sufficiently accurate approximant, so completion does not enlarge that value group. The final assertion follows when A~/P\widetilde A/P has dimension one. ◻

Lemma 4.24.3. NAG-002071 (Bounded expansions on pointed tubes). Let U\mathcal U be a flat finitely presented OO-scheme with a rational special-fibre point uu, and let U→AO1\mathcal U\to\mathbb A^1_O be étale at uu, sending uu to the special-fibre origin. Shrink U\mathcal U around uu and write zz for the target coordinate. Every integral function on its formal completion restricts to its tube as a power series g(z)=∑j≥0bjzj,bj∈O,∣z∣<1.g(z)=\sum_{j\geq0}b_jz^j,\qquad b_j\in O, \quad |z|<1. Reducing the coefficients gives the expansion of g‾\overline g in the completed local ring k[[z‾]]k[[\overline z]] of the special fibre at uu.

For an étale pointed chart (U,u)⟶(Spec⁡O[s,t]/(st−a),o),0<∣a∣<1,(\mathcal U,u)\longrightarrow \bigl(\operatorname{Spec}O[s,t]/(st-a),o\bigr), \qquad 0<|a|<1, with u,ou,o rational, the corresponding assertion is that the restriction to ]u[]u[ is represented by g=∑i,j≥0cijsitj,cij∈O,∣s∣<1,∣t∣<1,st=a.g=\sum_{i,j\geq0}c_{ij}s^it^j,\qquad c_{ij}\in O, \qquad |s|<1,\quad |t|<1,\quad st=a. On the normalization branch t=0t=0, the reduction of gg has completed expansion ∑i≥0c‾i0si\sum_{i\geq 0}\overline c_{i0}s^i; on s=0s=0, it has expansion ∑j≥0c‾0jtj\sum_{j\geq 0}\overline c_{0j}t^j.

Proof. After a Zariski shrinking preserving the entire tube, the square Jacobian presentation of [StacksAlg] (Tag 02GU) expresses the chart by a system F(Z,W)=0F(Z,W)=0 with coefficients in OO, where Z=zZ=z in the smooth case and Z=(s,t)Z=(s,t) in the nodal case. Include the inverses of all localized functions as extra variables and equations. The Jacobian with respect to WW is invertible at the selected residual point. In the nodal case take polynomial representatives of the coefficients before imposing st=ast=a.

At Z=0Z=0, the residual solution of F(0,W)=0F(0,W)=0 has invertible Jacobian. Lemma 4.15.1 gives a solution w0∈Onw_0\in O^n lifting it. There is a unique formal solution W(Z)∈O[[Z]]nW(Z)\in O[[Z]]^n with constant term w0w_0. Indeed, after coefficients of degrees less than mm have been chosen, the equations in degree mm are linear equations with coefficient matrix JWF(0,w0)∈GLn(O)J_WF(0,w_0)\in\mathrm{GL}_n(O); solve them for every monomial of degree mm. This recursive construction also proves uniqueness.

All coefficients lie in OO, so these series converge on every closed polydisc of radii strictly less than one. Their values reduce to the prescribed point. The uniqueness in Corollary 4.11.2 identifies this analytic solution with the inverse of the pointed tube map. Choose a polynomial representative for each regular function in the presentation, including the inverse variables. Substitution gives its bounded-coefficient expansion. For an element of the π\pi-adic completion, choose successively regular functions gng_n such that g=∑n≥0πngng=\sum_{n\geq 0}\pi^ng_n. Expand each gng_n in this way and sum the resulting series with the factors πn\pi^n. The coefficients converge in OO, uniformly on every closed smaller polydisc. On the nodal tube the imposed relation st=ast=a ensures equality with the original function, regardless of the polynomial representatives used.

Reduction commutes with this construction. The reduced recursive solution is the inverse of the pointed étale map on completed special-fibre local rings. In the smooth case this is k[[z]]k[[z]]. In the nodal case impose st=0st=0 and then restrict to either branch, giving the two stated expansions. Equivalently, these are the corresponding expansions on the normalization; normalization commutes with étale base change by [StacksAlg] (Tag 03GV). ◻

NAG-002072 · Open at full scope

Lemma 4.24.4. NAG-002072 (The first residual coefficient determines the slope). Let g(z)=∑j≥0bjzjg(z)=\sum_{j\geq 0}b_jz^j on ∣z∣<1|z|<1, with bj∈Ob_j\in O, and suppose its reduction as a formal series is nonzero of order dd. Then for all ρ<1\rho<1 sufficiently close to one, ∣g(ζ0,ρ)∣=ρd.|g(\zeta_{0,\rho})|=\rho^d. The same conclusion holds near the outer end of ∣a∣<∣s∣<1|a|<|s|<1 for a bounded double series g=∑i,j≥0cijsi(a/s)jg=\sum_{i,j\geq 0}c_{ij}s^i(a/s)^j, if ∑ic‾i0si\sum_i\overline c_{i0}s^i is nonzero of order dd.

Proof. In the disc case, ∣bd∣=1|b_d|=1 and ∣bj∣<1|b_j|<1 for j<dj<d. Choose ρ0<1\rho_0<1 so that ∣bj∣<ρ0d−j|b_j|<\rho_0^{d-j} for every j<dj<d. For ρ0<ρ<1\rho_0<\rho<1, the ddth term has Gauss size ρd\rho^d, every lower term has smaller size, and every higher term has size at most ρd+1\rho^{d+1}. The asserted norm follows.

For the annulus, regroup the convergent double series by the exponent of ss. The summability criterion 1.1.2, applied on every closed subannulus, justifies the regrouping. Its Laurent coefficients are bn=∑j≥0cn+j,jaj(n≥0),b−m=am∑i≥0ci,i+mai(m≥1).b_n=\sum_{j\geq0}c_{n+j,j}a^j\quad(n\geq0),\qquad b_{-m}=a^m\sum_{i\geq0}c_{i,i+m}a^i\quad(m\geq1). Consequently bn∈Ob_n\in O, b‾n=c‾n0\overline b_n=\overline c_{n0} for n≥0n\geq 0, and ∣b−m∣≤∣a∣m|b_{-m}|\leq|a|^m. Choose ρ0\rho_0 as in the disc case and increase it, if necessary, so that ∣a∣<ρ0d+1|a|<\rho_0^{d+1}. For ρ0<ρ<1\rho_0<\rho<1 all negative terms satisfy ∣b−m∣ρ−m≤(∣a∣/ρ)m≤∣a∣/ρ<ρd.|b_{-m}|\rho^{-m} \leq(|a|/\rho)^m\leq |a|/\rho<\rho^d. The positive terms are handled exactly as above, leaving the ddth term as the unique maximal term. ◻

Theorem 4.24.5. NAG-002073 (Residual order on a model branch). Let X\mathcal X be a flat finitely presented OO-curve with reduced nodal special fibre and smooth generic fibre. Assume that its formal generic fibre XX is Hausdorff. Let CC be a special-fibre component, x=xCx=x_C its component point, and ν:Cν→C\nu:C^\nu\to C its normalization. Let p∈Cν(k)p\in C^\nu(k) lie over either a smooth point of Xs\mathcal X_s, or a split node admitting a pointed chart as in Proposition 4.17.3. The pointed disc or annulus chart determines an outgoing Gauss segment γp:[0,ϵ)→X\gamma_p:[0,\epsilon)\to X, with γp(0)=x\gamma_p(0)=x and parameter τ=−log⁡∣z∣\tau=-\log|z|, where zz is the coordinate of the corresponding normalization branch and ∣z(x)∣=1|z(x)|=1.

Let ff be an analytic germ at xx with f(x)≠0f(x)\ne 0. Choose c∈K×c\in K^\times such that ∣cf(x)∣=1|cf(x)|=1, and let f~x∈k(C)×\widetilde f_x\in k(C)^\times be the residue of cf(x)cf(x) under Proposition 4.24.2. For all sufficiently small τ≥0\tau\geq 0, −log⁡∣f(γp(τ))∣=−log⁡∣f(x)∣+ord⁡p(f~x) τ.-\log|f(\gamma_p(\tau))| =-\log|f(x)|+\operatorname{ord}_p(\widetilde f_x)\,\tau. In particular the outgoing slope equals ord⁡p(f~x)\operatorname{ord}_p(\widetilde f_x), independently of the scalar cc and of the pointed chart.

Proof. We first treat an actual pointed chart on a formal affine neighborhood of the residual point. Its generic fibre is M(A)\mathcal M(A), with reduced integral model A0A_0. The Gauss segment has endpoint xx by Propositions 4.20.3 and 4.21.1. Choose P,Q∈AP,Q\in A with Q(x)≠0Q(x)\ne 0 such that the element r(x)=P(x)/Q(x)r(x)=P(x)/Q(x) of H(x)\mathcal H(x) satisfies ∣f(x)−r(x)∣<∣f(x)∣.|f(x)-r(x)|<|f(x)|. This is possible because the fraction field of A/ker⁡xA/\ker x is dense in H(x)\mathcal H(x). The strict inequality persists on a neighborhood of xx on which ff and P/QP/Q are analytic. The ultrametric inequality then gives ∣f∣=∣r∣|f|=|r| there. Moreover cf(x)cf(x) and cr(x)cr(x) have the same residue. It therefore suffices to prove the formula for P/QP/Q.

Apply the separator construction in the proof of Proposition 4.24.2 to PP and QQ, using the same function hh and the same exponent NN. For suitable cP,cQ∈K×c_P,c_Q\in K^\times, the functions P′=PhN/cP,Q′=QhN/cQP'=Ph^N/c_P,\qquad Q'=Qh^N/c_Q belong to A0A_0 and satisfy ∣P′(x)∣=∣Q′(x)∣=1|P'(x)|=|Q'(x)|=1. Their restrictions to CC are nonzero regular functions near the selected residual point. Write dP,dQd_P,d_Q for their orders at pp on CνC^\nu. By Lemmas 4.24.3 and 4.24.4, near the endpoint of the corresponding Gauss segment we have ∣P′(γp(τ))∣=e−dPτ,∣Q′(γp(τ))∣=e−dQτ.|P'(\gamma_p(\tau))|=e^{-d_P\tau},\qquad |Q'(\gamma_p(\tau))|=e^{-d_Q\tau}. In the nodal case use the branch coordinate that has absolute value one at xx; at the other end this exchanges ss and t=a/st=a/s. Since P/Q=(cP/cQ)P′/Q′P/Q=(c_P/c_Q)P'/Q', subtracting the two logarithms proves the displayed formula. The quotient of the two reductions is f~x\widetilde f_x up to a nonzero constant, so its order is dP−dQd_P-d_Q.

For a pointed étale diagram, pull the calculation back to its source U\mathcal U. The component point selected by the chosen branch maps to xx: its reduction is the corresponding generic point, whose analytic inverse image is unique by Proposition 4.12.2. Normalization commutes with this étale map by [StacksAlg] (Tag 03GV). On the normalized curves the local map at the rational branch point is therefore étale. Its fibre is a finite separable residue field by [StacksAlg] (Tag 02GL), so the upstairs maximal ideal is generated by the downstairs maximal ideal. A uniformizer downstairs is consequently a uniformizer upstairs. Thus the order of the pulled-back residual function is exactly ord⁡p(f~x)\operatorname{ord}_p(\widetilde f_x). The pointed tube map is an analytic isomorphism, so the calculated norm is the original norm along γp\gamma_p.

Multiplying cc by a scalar of absolute value one multiplies the residue by an element of k×k^\times, which has order zero. The right hand side is intrinsic to the normalized residual point; the same computation in any other pointed chart gives that same number. The logarithmic parameter is the annular length parameter of Proposition 4.19.1. ◻

Proposition 4.24.6. NAG-002074 (Slopes at separable residual points). Keep the curve model and component of Theorem 4.24.5. Let pp be a closed point of CνC^\nu with κ(p)/k\kappa(p)/k separable. Choose a finite Galois residue-field lift L/KL/K which makes the points above pp rational and supplies pointed charts at any node under pp. Let p′p' lie above pp on the normalization of a component C′C' of the extended special fibre, let x′x' be its component point, and let γp′\gamma_{p'} be an outgoing Gauss segment at x′x'. Writing ϖ:XL→X\varpi:X_L\to X for projection, the projected segment γ=ϖ∘γp′\gamma=\varpi\circ\gamma_{p'} is nonconstant and injective on a sufficiently short initial interval. For every analytic germ ff at xx with f(x)≠0f(x)\ne 0, −log⁡∣f(γ(τ))∣=−log⁡∣f(x)∣+ord⁡p(f~x) τ(0≤τ≪1).-\log|f(\gamma(\tau))| =-\log|f(x)|+\operatorname{ord}_{p}(\widetilde f_x)\,\tau \qquad(0\leq\tau\ll1). The coefficient on the right is independent of LL, of p′p', and of the pointed chart. The parameter uses the original absolute value on KK, extended to LL; no factor [L:K][L:K] occurs in this slope.

Proof. Take a finite Galois residue extension containing the normal closure of κ(p)/k\kappa(p)/k. If pp lies over a node, enlarge it to include the finite residue extension used in Theorem 4.18.2. Its lift exists by Lemma 4.18.1. The algebraic and formal base changes agree with the analytic scalar extension by Lemma 4.22.1. Their special fibres are obtained by tensoring with l=L~l=\widetilde L, and remain reduced.

The finite separable morphism Spec⁡l→Spec⁡k\operatorname{Spec}l\to\operatorname{Spec}k is étale. By [StacksAlg] (Tag 03GV), the normalization of the extended residual curve is the base change of the original normalization. The local map at p′p' over pp is therefore étale. By [StacksAlg] (Tag 02GL), its maximal ideal is generated by the image of the maximal ideal at pp. Both local rings are discrete valuation rings, by [StacksAlg] (Tag 0B8Y). Consequently ord⁡p′(u∣C′)=ord⁡p(u)(u∈k(C)×).\operatorname{ord}_{p'}(u|_{C'}) =\operatorname{ord}_{p}(u)\qquad(u\in k(C)^\times). The generic point of C′C' maps to that of CC, so specialization and the uniqueness of component points give ϖ(x′)=x\varpi(x')=x. The residue-field isomorphisms in Proposition 4.24.2 commute with these maps: they are induced by evaluating integral model functions and then taking residues. Thus the normalized residue of the pullback of ff is the pullback of f~x\widetilde f_x. Apply Theorem 4.24.5 at x′,p′x',p' and use the displayed order equality. Evaluation of a pulled-back function has exactly its original absolute value. This proves the formula with the stated parameter and coefficient.

To see that the projected path is nonconstant and has no initial self-identifications, choose a uniformizer u∈k(C)×u\in k(C)^\times at pp. Write u=P‾/Q‾u=\overline P/\overline Q in the fraction field of a residual affine chart, with both numerator and denominator nonzero on CC, and lift them to integral functions P,QP,Q. Then r=P/Qr=P/Q is analytic near xx, ∣r(x)∣=1|r(x)|=1, and its normalized residue is uu. The already proved formula for rr gives ∣r(γ(τ))∣=e−τ|r(\gamma(\tau))|=e^{-\tau} on a sufficiently short interval. These values distinguish its parameters. ◻

Corollary 4.24.7. NAG-002075 (Residual-degree weighted balancing). Assume that kk is perfect and that the residual component CC is proper over kk. For a nonzero analytic germ ff at its component point xx, with f(x)≠0f(x)\ne 0, index the model branches at xx by the closed points pp of CνC^\nu and write dpF(x)d_pF(x) for the slopes of F=−log⁡∣f∣F=-\log|f| constructed above. Then only finitely many of these slopes are nonzero and ∑p∈Cν closed[κ(p):k] dpF(x)=0.\sum_{p\in C^\nu\ {\rm closed}} [\kappa(p):k]\,d_pF(x)=0. If kk is algebraically closed, every weight is one. For a general perfect residue field, the weights are the degrees over the original kk, not over a splitting field.

Proof. The normalization is finite by [StacksAlg] (Tag 0BXR). Hence CνC^\nu is a proper normal integral curve, with discrete valuation local rings at closed points by [StacksAlg] (Tag 0B8Y). Perfection makes all its finite residue extensions separable, so Proposition 4.24.6 applies to every closed point and gives dpF(x)=ord⁡p(f~x).d_pF(x)=\operatorname{ord}_{p}(\widetilde f_x). A rational function on a Noetherian integral curve is a unit on a dense open subset, whose complement is a finite set of closed points. This proves the finiteness assertion. The degree-zero formula for its principal divisor is exactly [StacksAlg] (Tag 02RU), applied over the original field kk to the proper curve CνC^\nu. It gives the stated weighted sum. ◻

References. The residual-curve formulation follows the organization of [BPR13] (§5.13 and Theorem 5.15). The analytic order comparison above is proved from bounded pointed-chart expansions and the internal annulus norm calculations. The algebraic balancing step uses the proved Stacks principal-divisor degree theorem, with the original residue-field degrees retained.

4.25 Residual branches as analytic directions

Lemma 4.25.1. NAG-002066 (The radial germ of an open disc). Let DD be an open unit disc with coordinate TT, and let ζr\zeta_r, 0<r<10<r<1, be its centered Gauss points. An analytic isomorphism between two such discs sends ζr\zeta_r to the centered radius-rr Gauss point for all rr sufficiently close to 11. On a sufficiently thin outer collar it also preserves the function ∣T∣|T|. These assertions hold for semilinear isometric isomorphisms and commute with complete isometric ground-field extension. The open disc and each collar {r<∣T∣<1}\{r<|T|<1\} are path connected.

Proof. First suppose that the ground valuation is nontrivial. Write an isomorphism as g(T)=∑anTng(T)=\sum a_nT^n. The proof of Theorem 4.21.2 gives ∣a0∣<1|a_0|<1, ∣a1∣=1|a_1|=1, and ∣an∣≤1|a_n|\leq 1 for n≥2n\geq 2. For ∣a0∣<∣T(x)∣<1|a_0|<|T(x)|<1, the linear term uniquely dominates all others, so ∣g(x)∣=∣T(x)∣|g(x)|=|T(x)|. The same statement for the inverse makes the corresponding collars identical after increasing their lower radius if necessary. The homotopy identity gives g(ζr)=g(HD(0,r))=HD(g(0),r)=ζr(r>∣a0∣).g(\zeta_r)=g(H_D(0,r))=H_D(g(0),r)=\zeta_r \qquad(r>|a_0|). The last equality is invariance of the radius-rr Gauss seminorm under translation by an element of absolute value at most rr. It follows directly by applying the ultrametric inequality first to translation and then to its inverse. This argument also gives HD(x,r)=ζrH_D(x,r)=\zeta_r whenever ∣T(x)∣≤r|T(x)|\leq r. Hence any two points of DD can be joined, using HDH_D up to a common time r<1r<1, to the same ζr\zeta_r. For a collar, the radial annulus homotopy joins each point to the Gauss point of the same radius, and the Gauss interval joins any two of these points inside the collar. These are continuous paths by Proposition 4.3.1. All formulas preserve the absolute value under semilinear isometries and agree on original polynomials after scalar extension, proving the remaining assertions.

It remains to treat a trivially valued KK. Choose 0<η<10<\eta<1. Then (η)(\eta) is free over KK, and Lemma 1.16.1 supplies a complete isometric extension L=KηL=K_\eta with a Laurent parameter of absolute value η\eta. The valuation on LL is nontrivial. An isometric automorphism of KK extends coefficient by coefficient to LL, fixing that parameter. This extension preserves the weighted Laurent norm and its inverse is constructed in the same way.

For either given coordinate, take real radii si↑1s_i\uparrow1, with 0<si<10<s_i<1. The interiors of the closed discs of radii sis_i cover the open disc. In the actual completed tensor presentation their algebras satisfy K⟨si−1T⟩⊗^KL=L⟨si−1T⟩.K\langle s_i^{-1}T\rangle\widehat\otimes_K L = L\langle s_i^{-1}T\rangle . Indeed, the weighted Laurent coefficient description of Theorem 1.16.2 for the free radius η\eta on the left is the weighted coefficient description for the same two indices on the right. In both descriptions the norm is the supremum with weights sinηjs_i^n\eta^j and the weighted coefficient family tends to zero. The identifications commute with restriction to smaller discs and the radii are cofinal at every point with ∣T∣<1|T|<1 over both fields.

Apply Theorem 2.40.12 to each open disc with its given coordinate. The scalar projections are continuous and surjective and commute with the given isomorphism and its extension. For a semilinear isomorphism, use the coefficient automorphism of LL just described. Since projection restricts a seminorm to the original coefficients, the function ∣T∣\lvert T\rvert pulls back to the same coordinate norm. The inverse image of each collar q<∣T∣<1q<\lvert T\rvert<1 is therefore the corresponding collar. The radius-rr centered Gauss point over LL projects to the radius-rr centered Gauss point over KK, by its formula on K[T]K[T].

Apply the nontrivially valued case to the lifted isomorphism. It identifies the two collars and preserves their coordinate norms as soon as qq bounds the absolute values at zero of the coordinate series of the isomorphism and its inverse. Both constants lie in KK and have norm less than one, so they vanish in the trivially valued case. Surjectivity on each collar, and the commutative square with the scalar projections, give these assertions downstairs. Applying the same square to the centered Gauss points gives their asserted images. For any two points of the disc or of a collar, choose their two lifts individually by surjectivity and join the lifts upstairs. The projection of that path joins the given endpoints. This uses only the surjectivity of the scalar projection. For any further complete isometric scalar extension the same projection identities on coordinate norms and on the Gauss seminorms follow by restriction of seminorms. These identities give scalar compatibility for the asserted collars and centered points. ◻

Proposition 4.25.2. NAG-002067 (The canonical germ of a separable smooth tube). In the setting of Proposition 4.22.2, write Dq=]q[D_q=]q[ and Cq=Dq∪{v}C_q=D_q\cup\{v\}. There is a canonical germ of an embedded Gauss ray approaching vv from DqD_q, parametrized near vv by ℓ=−log⁡r≥0\ell=-\log r\geq 0. There are also path-connected open collars Bq(r)⊂DqB_q(r)\subset D_q, for r<1r<1 sufficiently close to 11, such that {v}∪Bq(r)\{v\}\cup B_q(r) form a neighborhood basis of vv in CqC_q. The tube DqD_q is path connected. The germ and its parameter are independent of the finite residue-field lift and disc coordinates used to construct them.

Proof. Use a finite Galois residue-field lift L/KL/K splitting qq, as in Proposition 4.22.2. Write G=Gal⁡(L/K)G=\operatorname{Gal}(L/K) and p−1(Dq)=∐jDjp^{-1}(D_q)=\coprod_j D_j, where DjD_j is an open unit disc, and Cj=Dj∪{vj}C_j=D_j\cup\{v_j\}. The group GG acts transitively on the residual points qjq_j: they are the embeddings of κ(q)\kappa(q) into the chosen normal residue extension, and residue-field Galois automorphisms lift by Lemma 4.18.1. Thus p(Dj)=Dqp(D_j)=D_q and p(Cj)=Cqp(C_j)=C_q for each jj. In particular DqD_q is path connected by the preceding lemma.

Choose a coordinate TjT_j on every DjD_j. There are only finitely many transition isomorphisms induced by GG. Lemma 4.25.1 therefore supplies a common r0<1r_0<1 such that, for all r>r0r>r_0, each transition preserves the entire collar ∣Tj∣>r|T_j|>r and the centered Gauss point ζj,r\zeta_{j,r}. Consequently Bq(r)=p({∣Tj∣>r}),γq(r)=p(ζj,r)B_q(r)=p(\{|T_j|>r\}),\qquad \gamma_q(r)=p(\zeta_{j,r}) are independent of jj. The union of the collars upstairs is GG-invariant, and its complement after adjoining all vjv_j is the finite union of the compact closed subdiscs ∣Tj∣≤r|T_j|\leq r. By the quotient identification of Theorem 4.9.2, {v}∪Bq(r)\{v\}\cup B_q(r) is open in CqC_q. Each Bq(r)B_q(r) is a continuous image of an open annulus, hence path connected. For any neighborhood NN of vv in CqC_q, its inverse image contains all vjv_j. The rational tube neighborhood basis of Lemma 4.21.1 puts all their sufficiently thin collars inside p−1(N)p^{-1}(N). There are finitely many jj, so one common rr works. This proves the neighborhood-basis assertion.

Set γq(1)=v\gamma_q(1)=v. The endpoint result makes γq:(r0,1]→Cq\gamma_q:(r_0,1]\to C_q continuous. If two of its values with parameters greater than r0r_0 coincide, the orbit description of analytic fibres in Theorem 4.9.2 sends one chosen Gauss point to the other. Every such group element preserves its radius on these collars, so the two parameters agree. The function ∣Tj∣|T_j| descends to a continuous function on Bq(r0)B_q(r_0) and extends by value 11 at vv, by the collar neighborhood basis. Its restriction is the inverse parameter of γq\gamma_q, proving that the ray is embedded.

Changing the finitely many disc coordinates preserves these radii and rays after increasing r0r_0, by Lemma 4.25.1. Two residue-field lifts embed into a common such lift as in the proof of Proposition 4.22.2. Gauss points and collars commute with scalar projection, so the germs agree in that common comparison. The length parameter is ℓ=−log⁡r\ell=-\log r throughout; the residue extension does not alter the fixed absolute value. ◻

Theorem 4.25.3. NAG-002068 (The residual branch star). Assume Theorem 4.23.3. Let CC be a special-fibre component, let v=vCv=v_C, and let CνC^\nu be its normalization. There is an ordinary open neighborhood WvW_v of vv whose punctured connected components are in natural bijection with the closed points of CνC^\nu. At a smooth point of the special fibre the corresponding component is its whole open tube. At a point above a node it is a thin annular collar of the selected node branch. A self-node contributes two distinct components.

These components remain naturally indexed by the same closed points in a cofinal system of neighborhoods of vv, obtained by making the finitely many nodal collars thinner and replacing finitely many smooth tubes by outer collars. Thus the germs of punctured-neighborhood components, or outgoing directions at vv, identify canonically with the closed points of CνC^\nu. Each has the coordinate- independent Gauss-ray representative described above.

Proof. Use the finite closed cover of XX by the component stars and completed nodal tubes in the proof of Theorem 4.23.3. Include the whole star ACA_C. For every node incident to CC, include a sufficiently thin open radial collar at each of its ends attached to vv, together with vv. For a self-node choose two disjoint collars inside its open annulus. Include no other vertices or component stars. Call the union WvW_v.

Its intersection with each component star is either the whole ACA_C or empty. Its intersection with a completed nodal tube is an open endpoint neighborhood, the union of two disjoint endpoint neighborhoods after their common vertex is identified, or empty. These sets are relatively open by Proposition 4.20.3 and the compact quotient in Corollary 4.20.4. A subset whose intersections with every member of a finite closed cover are relatively open is open: its complement is a finite union of closed subsets. Hence WvW_v is open.

After deleting vv, its pieces are precisely the smooth tubes on CC and the chosen annular collars. Each piece is open, and its complement is the union of the other open pieces, so it is also closed in Wv∖{v}W_v\setminus\{v\}. Each is path connected by Proposition 4.25.2 and Lemma 4.25.1. They are therefore exactly the connected components. Lemma 4.24.1 identifies their labels with all closed points of CνC^\nu, including two distinct normalization points for a self-node.

Finally let NN be any neighborhood of vv in XX. Lemma 4.23.1 puts all but finitely many smooth tubes on CC inside NN. For each of the remaining tubes use the collar neighborhood basis in Proposition 4.25.2; for each incident nodal end use the full radial tails of Proposition 4.20.3. Shrinking these finitely many collars yields a neighborhood contained in N∩WvN\cap W_v. It is open by the same finite-cover argument, since deleting a compact inner portion from finitely many smooth tubes is open on ACA_C. Its punctured pieces are again path connected, with unchanged labels. This proves cofinality and makes all transition maps on component labels the identity. The claimed direction correspondence and Gauss-ray representatives follow. ◻

4.26 Marked rays and rational functions

Proposition 4.26.1. NAG-002076 (A punctured smooth tube). Let X/O\mathcal X/O be a curve model, smooth of relative dimension one at a rational special-fibre point qq. Let D=]q[D=]q[, let vv be its component point, and let a∈D(K)a\in D(K). There is a pointed disc coordinate zz with z(a)=0z(a)=0, identifying DD with ∣z∣<1|z|<1. The ray Ra={v}∪{ζ0,ρ:0<ρ<1},τ=−log⁡ρ∈[0,∞)R_a=\{v\}\cup\{\zeta_{0,\rho}:0<\rho<1\}, \qquad \tau=-\log\rho\in[0,\infty) is intrinsic to the pointed tube (D,a)(D,a). The radial annulus homotopy on D∖{a}D\setminus\{a\} extends by fixing vv to a jointly continuous strong deformation retraction of (D∪{v})∖{a}(D\cup\{v\})\setminus\{a\} onto RaR_a.

Proof. Take the pointed smooth coordinate supplied by Proposition 4.17.2. Its value at aa belongs to K∘∘K^{\circ\circ}, so subtracting that value retains the unit-disc coordinate and puts aa at zero. The complement is the standard open annulus 0<∣z∣<10<|z|<1. Corollary 4.19.1 supplies its radial homotopy. It preserves ∣z∣|z| at every time. By Lemma 4.21.1, every net in DD whose radii tend to one converges to vv, uniformly over each outer radial tail. This proves joint continuity at all (v,u)(v,u), 0≤u≤10\leq u\leq 1, after setting H(v,u)=vH(v,u)=v.

A change between two centered disc coordinates restricts to an isomorphism of 0<∣z∣<10<|z|<1. The radius transformation in Corollary 4.19.1 must be ρ↦ρ\rho\mapsto\rho: it has the form ∣c∣ρ±1|c|\rho^{\pm 1} and maps (0,1)(0,1) onto itself. That corollary identifies both the Gauss ray and the entire homotopy. The logarithmic parameter is therefore unchanged. ◻

Theorem 4.26.2. NAG-002077 (The skeleton of a marked model). Let X\mathcal X satisfy Theorem 4.23.3. Choose finitely many distinct KK-points a1,…,ama_1,\ldots,a_m whose specializations q1,…,qmq_1,\ldots,q_m are distinct rational smooth points of the special fibre. Put P={a1,…,am}P=\{a_1,\ldots,a_m\} and define Σ(X,P)=Σ(X)∪⋃i=1mRai.\Sigma(\mathcal X,P)=\Sigma(\mathcal X)\cup \bigcup_{i=1}^m R_{a_i}. This is the finite model graph with one infinite metric ray attached at the component point of each qiq_i. It is a strong deformation retract of X∖PX\setminus P, canonically relative to the marked model. Its retraction agrees with the model retraction outside the marked smooth tubes and with y↦ζ0,∣zi(y)∣y\mapsto\zeta_{0,|z_i(y)|} on their punctured interiors.

Proof. Let Di=]qi[D_i=]q_i[ and Ci=Di∪{vi}C_i=D_i\cup\{v_i\}. Each CiC_i is closed in XX by Lemma 4.21.1. The space X∖PX\setminus P has the finite closed cover consisting of X∖⋃iDi,Ci∖{ai}(1≤i≤m).X\setminus\bigcup_iD_i,\qquad C_i\setminus\{a_i\} \quad(1\leq i\leq m). On the first set restrict the model homotopy. That set is preserved because each unmarked smooth tube, each nodal tube, and every vertex are preserved. On Ci∖{ai}C_i\setminus\{a_i\} use Proposition 4.26.1. The intersections consist only of fixed vertices, so finite closed pasting, also after product with [0,1][0,1], gives a jointly continuous homotopy. Its initial map is the identity, its final image is the stated graph with rays, and it fixes that image pointwise. These are the strong retraction identities.

The coordinate τ=−log⁡∣zi∣\tau=-\log|z_i| gives the ray its ordinary interval topology and length. As τ↓0\tau\downarrow 0 the ray tends to viv_i; as τ→∞\tau\to\infty it tends to aia_i in XX, by the disc seminorm formula. Adjoin each aia_i as the other endpoint of its compactified ray. The resulting finite compact gluing maps continuously and injectively into the Hausdorff space XX, hence is homeomorphic to its image. Removing the aia_i gives precisely the asserted finite graph with infinite rays. The local coordinate independence in Proposition 4.26.1 and the model homotopy’s invariance prove the final assertion. ◻

Theorem 4.26.3. NAG-002078 (Factorization and orders at the marked ends). Assume in addition that X\mathcal X is projective over OO, that kk is perfect, and that its generic fibre is a smooth integral curve XKX_K. Identify its analytic generic fibre with XKanX_K^{\mathrm{an}} by Proposition 4.13.2. Let f∈K(XK)×f\in K(X_K)^\times, and suppose the support of div⁡(f)\operatorname{div}(f) is contained in the marked set PP. Writing rPr_P for the retraction onto Σ(X,P)\Sigma(\mathcal X,P), one has F=−log⁡∣f∣=F∘rPon XKan∖P.F=-\log|f|=F\circ r_P \qquad\hbox{on }X_K^{\mathrm{an}}\setminus P. The restriction of FF is affine with integral slope on each finite edge and on each marked ray. On the ray toward aia_i, its outgoing slope is ord⁡ai(f)\operatorname{ord}_{a_i}(f). At every component vertex it satisfies the residual-degree weighted balancing identity of Corollary 4.24.7.

Proof. The rational function is an algebraic unit off PP, because the local rings of the smooth curve are discrete valuation rings, [StacksAlg] (Tag 0B8Y). Analytification therefore makes it an analytic unit on XKan∖PX_K^{\mathrm{an}}\setminus P. On every nodal tube, Corollary 4.10.3 gives ∣f∣=∣c∣∣z∣n|f|=|c||z|^n with n∈Zn\in\mathbb Z. Thus FF factors through the annular retraction and is affine of integral slope on its edge.

On an unmarked smooth tube with rational residue point, the same corollary makes ∣f∣|f| constant. For a separable nonrational residue point, split its residue field as in Proposition 4.22.2. The pullback of ff has constant norm on each resulting disc, equal by endpoint continuity to its value at the corresponding component point. Those component points all map to the original vertex, so all these constants equal ∣f(v)∣|f(v)|. Surjectivity of projection shows that ∣f∣|f| has this constant value on the original tube.

Fix a marked tube. Its coordinate ziz_i can be chosen from a smooth algebraic chart and centered at aia_i, as in Proposition 4.26.1. The generic-fibre chart is étale at aia_i and has residue field KK. By [StacksAlg] (Tag 02GU(6)), ziz_i generates the maximal ideal of the curve’s local ring at aia_i. It is therefore a uniformizer. Put mi=ord⁡ai(f)m_i=\operatorname{ord}_{a_i}(f). The quotient gi=f/zimig_i=f/z_i^{m_i} is an algebraic unit near aia_i; on the remainder of its tube it is an analytic unit, because ziz_i has no other zero there and ff has no other zero or pole there. These expressions agree on their overlap, so gig_i is an analytic unit on the whole tube. Hence ∣f(y)∣=∣gi(ai)∣ ∣zi(y)∣mi,F∣Rai(τ)=miτ−log⁡∣gi(ai)∣.|f(y)|=|g_i(a_i)|\,|z_i(y)|^{m_i}, \qquad F|_{R_{a_i}}(\tau) =m_i\tau-\log|g_i(a_i)|. This proves both the factorization on the marked tube and the exact order at its end. The displayed local formulas cover X∖PX\setminus P and agree at the component vertices by continuity, proving the global factorization. Each residual component is proper, since it is a closed component of the projective special fibre. Thus Corollary 4.24.7 applies at every vertex and proves the last assertion. ◻

4.27 Inserting vertices by explicit blowups

Let KK be complete and nontrivially valued. We write O=K∘O=K^\circ, m=K∘∘\mathfrak m=K^{\circ\circ} and k=O/mk=O/\mathfrak m. A nodal curve model in this section is a flat finitely presented OO-scheme with smooth generic fibre, reduced nodal special fibre, and Hausdorff formal generic fibre.

Lemma 4.27.1. NAG-002099 (A symmetric subdivision of a nodal annulus). Let b,λ∈m∖{0}b,\lambda\in\mathfrak m\setminus\{0\} with ∣b∣≤∣λ∣2|b|\leq|\lambda|^2, and put c=b/λ2∈Oc=b/\lambda^2\in O. The blowup of (λ,s,t)(\lambda,s,t) in Spec⁡O[s,t]/(st−b)\operatorname{Spec}O[s,t]/(st-b) is covered by the three flat finitely presented charts generatorchart ringmap to the original chartλO[u,v]/(uv−c)s=λu,t=λvsO[s,w]/(sw−λ)t=λcwtO[t,w′]/(tw′−λ)s=λcw′.\begin{array}{c|c|c} \text{generator}& \text{chart ring}& \text{map to the original chart}\\ \hline \lambda&O[u,v]/(uv-c)&s=\lambda u,\quad t=\lambda v\\ s&O[s,w]/(sw-\lambda)&t=\lambda c w\\ t&O[t,w']/(tw'-\lambda)&s=\lambda c w'. \end{array} Its formal completion has the same analytic generic fibre as the original model. On the original annular skeleton it inserts the radii ∣s∣=∣λ∣|s|=|\lambda| and ∣s∣=∣b∣/∣λ∣|s|=|b|/|\lambda|. These are two new component points if ∣b∣<∣λ∣2|b|<|\lambda|^2, and one if ∣b∣=∣λ∣2|b|=|\lambda|^2. The resulting edge lengths are, respectively, −log⁡∣λ∣,log⁡∣λ∣2∣b∣,−log⁡∣λ∣,or−log⁡∣λ∣,−log⁡∣λ∣.-\log|\lambda|,\quad \log\frac{|\lambda|^2}{|b|},\quad -\log|\lambda|, \qquad\text{or}\qquad -\log|\lambda|,\quad-\log|\lambda|.

Proof. Set A=O[s,t]/(st−b)A=O[s,t]/(st-b). Its normal form with monomials 1,si,tj1,s^i,t^j for i,j>0i,j>0 embeds AA into K[s,s−1]K[s,s^{-1}]. The affine blowup chart for λ\lambda is A[s/λ,t/λ]⊂A[1/λ]A[s/\lambda,t/\lambda]\subset A[1/\lambda]. It is exactly O[u,v]/(uv−c)O[u,v]/(uv-c): the displayed map is surjective onto that subring, and its injectivity follows from the same normal form, now with u=s/λu=s/\lambda and v=c/uv=c/u.

For the ss-chart, put w=λ/sw=\lambda/s. Inside A[1/s]A[1/s] one has t=bs=λcw,ts=cw2.t=\frac{b}{s}=\lambda c w,\qquad \frac{t}{s}=c w^2 . Consequently A[λ/s,t/s]A[\lambda/s,t/s] equals the subring generated by s,ws,w, with sole relation sw=λsw=\lambda. The normal form again proves injectivity. Interchanging s,ts,t gives the third chart. Each chart is a free OO-module with the indicated normal-form basis, so is flat; its displayed presentation is finite. These computations identify the actual blowup subrings, not unsaturated presentations with extra vertical torsion. Their completions are the admissible charts of Theorem 3.15.1.

If ∣c∣<1|c|<1, the middle special chart is uv=0uv=0. The two outer special charts are sw=0sw=0 and tw′=0tw'=0. On the first overlap u=w−1u=w^{-1}; on the second v=(w′)−1v=(w')^{-1}. Thus the two new components are projective lines, joined to each other and to the two old branches in a chain. If ∣c∣=1|c|=1, the middle special chart is uv=cˉuv=\bar c, and its two ends glue to the two outer affine pieces to give a single projective line between the old branches. The special fibres are reduced and nodal in both cases.

Theorem 3.15.2 identifies the completed generic charts with their rational domains. Lemma 3.20.3 then identifies the new component points with the stated Gauss radii. In the strict-inequality case the middle annulus has parameter cc; the outer annuli have parameter λ\lambda. In the equality case the middle generic chart is a circle. The logarithmic length formulas follow, and their sum is −log⁡∣b∣-\log|b|. ◻

Lemma 4.27.2. NAG-002100 (An intrinsic centre at a split node). Assume KK is algebraically closed. Let qq be a node of a nodal curve model X\mathcal X. Choose a pointed étale chart with equation st=bst=b, as in Theorem 4.15.3. For each ρ∈∣K×∣\rho\in|K^\times| with ∣b∣<ρ<1|b|<\rho<1 there is an admissible blowup of X^\widehat{\mathcal X}, induced by an algebraic blowup supported at qq, whose new nodal model contains the two annular radii ρ\rho and ∣b∣/ρ|b|/\rho as component points, counted only once if they coincide. All old component points are retained. A projective model remains projective.

Proof. Shrink an algebraic neighborhood UU of qq so that qq is its only nonsmooth special point. Set J=Fitt⁡1ΩU/O.\mathcal J=\operatorname{Fitt}_1\Omega_{U/O}. The module of differentials is finitely presented, and J\mathcal J is a finite-type quasi-coherent ideal [StacksAlg] (Tag 0C3E). For an étale map, differentials pull back isomorphically [StacksAlg] (Tag 04BX), and Fitting ideals commute with pullback [StacksAlg] (Tag 0C3D). The standard nodal presentation O[s,t]/(st−b)→ (t,s) (O[s,t]/(st−b))⊕2⟶Ω(O[s,t]/(st−b))/O⟶0O[s,t]/(st-b) \xrightarrow{\ (t,s)\ } \bigl(O[s,t]/(st-b)\bigr)^{\oplus2} \longrightarrow\Omega_{(O[s,t]/(st-b))/O}\longrightarrow0 therefore identifies the pullback of J\mathcal J with (s,t)(s,t) in every chosen pointed nodal chart.

Choose λ∈K×\lambda\in K^\times with ∣λ∣=max⁡(ρ,∣b∣/ρ)<1|\lambda|=\max(\rho,|b|/\rho)<1. Then ∣b∣≤∣λ∣2|b|\leq|\lambda|^2. On UU use the ideal I=(λ)+J\mathcal I=(\lambda)+\mathcal J. It is the unit ideal on U∖{q}U\setminus\{q\}: on the generic fibre λ\lambda is invertible, and at a smooth special point the differential module is free of rank one, so J\mathcal J is the unit ideal. Hence I\mathcal I glues with the unit ideal on X∖{q}\mathcal X\setminus\{q\}. This gives a global finite-type ideal supported at qq and containing (λ)(\lambda). A power of any fixed topologically nilpotent π\pi lies in (λ)(\lambda), so its completed ideal is open.

Blowup commutes with flat, in particular étale, base change [StacksAlg] (Tag 0805). Thus its pullback to the auxiliary pointed chart is exactly the blowup of (λ,s,t)(\lambda,s,t) in Lemma 4.27.1. This use of the Fitting ideal makes the centre descend even when both local branches belong to the same global component. The chosen auxiliary neighborhood can be shrunk so its fibre over qq is just the specified rational point. Its pullback over the exceptional fibre is consequently an isomorphism on that fibre. The new components and their attachment points are therefore precisely those computed in the standard charts.

The explicit charts are flat with reduced nodal special fibres and smooth generic fibres. These properties descend through the indicated étale neighborhoods; flatness is fpqc local on the source [StacksAlg] (Tag 036K), and local finite presentation is fppf local on the source [StacksAlg] (Tag 036N). The blowup is quasi-compact and separated over the original finite-presentation model, hence the new model is finitely presented. Its formal generic fibre is unchanged by Lemma 3.17.3, so remains Hausdorff. Away from qq the model is unchanged, including each old component generic point. The chart radii and lengths are those of Lemma 4.27.1. Finally the blowup is projective [StacksAlg] (Tag 02NS); if X\mathcal X is projective over OO, the composite is projective by [StacksAlg] (Tag 0C4P), since Spec⁡O\operatorname{Spec}O is quasi-compact and quasi-separated. ◻

Lemma 4.27.3. NAG-002101 (Insertion inside a smooth residual tube). Let qq be a rational smooth special point of a nodal curve model X\mathcal X, and let zz be the open-disc coordinate supplied by an algebraic smooth chart in Corollary 4.17.2. If a,λ∈Ka,\lambda\in K satisfy ∣a∣<1,0<∣λ∣<1,|a|<1,\qquad 0<|\lambda|<1, then an algebraic blowup supported at qq gives a new nodal model containing ζa,∣λ∣\zeta_{a,|\lambda|} as a component point. It retains all old component points and adds one projective-line special component, attached at the old component through qq. Its new edge has length −log⁡∣λ∣-\log|\lambda|. Projectivity of X/O\mathcal X/O is preserved.

Proof. Translate the algebraic smooth coordinate by a lift of its residue value at qq, so that its tube coordinate is zz. Shrink its Zariski chart until qq is the only special point with zˉ=0\bar z=0; the fibre of an étale morphism over that point is finite on the chosen affine chart. The ideal (λ,z−a)(\lambda,z-a) is then the unit ideal off qq. Glue it with the unit ideal on X∖{q}\mathcal X\setminus\{q\}. It is finite type and its formal completion is open.

Flat base change of the algebraic blowup [StacksAlg] (Tag 0805) reduces the calculation to Corollary 3.20.4, with T=z−aT=z-a and scale λ\lambda. The charts are T=λu,O[u],andO[T,v]/(Tv−λ).T=\lambda u,\quad O[u], \qquad\text{and}\qquad O[T,v]/(Tv-\lambda). Their completions, exceptional projective line, reduced nodal special fibre and generic-fibre identification are exactly those proved there. Pullback to the chosen étale chart introduces no extra exceptional components: its fibre over the selected rational point is a single rational point after the preceding shrinking. The exceptional component point is the Gauss point of ∣z−a∣≤∣λ∣|z-a|\leq|\lambda|, and its attaching annulus is ∣λ∣<∣z−a∣<1|\lambda|<|z-a|<1. The old model is unchanged away from qq. Flatness and finite presentation follow as in Lemma 4.27.2, while [StacksAlg] (Tags 02NS and 0C4P) preserves projectivity. ◻

Theorem 4.27.4. NAG-002102 (Realization of finitely many type II points). Assume KK is algebraically closed, complete and nontrivially valued. Let X\mathcal X be a nodal curve model and put X=X^ηX=\widehat{\mathcal X}_\eta. For every finite set SS of type II points of XX, a finite composition of admissible blowups of X^\widehat{\mathcal X}, induced by algebraic blowups of X\mathcal X, yields a nodal curve model whose component vertex set contains SS and all the old component vertices. If X\mathcal X is projective over OO, the new model is projective over OO.

Proof. By the tube decomposition in Theorem 4.23.3, a point not already a component vertex belongs to a smooth open-disc tube or to a nodal open-annulus tube. In either chart the exact classification of Theorem 4.1.1 represents a type II point as ζa,r\zeta_{a,r} with r∈∣K×∣r\in|K^\times|.

In a smooth tube, membership in the open unit disc gives ∣a∣<1|a|<1 and r<1r<1. Choose ∣λ∣=r|\lambda|=r and apply Lemma 4.27.3. In a nodal tube ∣b∣<∣s∣<1|b|<|s|<1, first suppose that the point is radial, namely ζ0,ρ\zeta_{0,\rho}. Apply Lemma 4.27.2 at radius ρ\rho.

For a nonradial point ζa,r\zeta_{a,r} in that annulus one has r<∣a∣r<|a|; otherwise the disc is also centred at zero. Here ∣b∣<∣a∣<1|b|<|a|<1. Insert the radial point of radius ∣a∣|a| first. In the new component its residue class is a nonzero finite value of the scaled coordinate, hence is a smooth point rather than one of the two attaching nodes. The point ζa,r\zeta_{a,r} now lies in that smooth residual tube, so a second application of Lemma 4.27.3 inserts it. Thus each desired point is inserted in at most two steps, possibly with one extra reflected annular vertex. Applying this procedure successively to SS terminates; every step preserves the previously obtained vertices and the stated model properties. ◻

Theorem 4.27.5. NAG-002103 (A model adapted to finitely many marked points). Under the hypotheses of Theorem 4.27.4, let P⊂X(K)P\subset X(K) be finite. There is a model obtained by finitely many of the preceding admissible blowups such that the points of PP specialize to pairwise distinct smooth closed points of the special fibre. The construction can retain any finite set of previously specified type II vertices. It preserves projectivity.

Proof. A KK-rational analytic point cannot specialize to a component generic point, since Lemma 4.24.2 gives the latter a residual function field of transcendence degree one, whereas the rational point has completed residue field KK. Thus every point of PP lies in a closed-point tube. If a marked point belongs to a nodal tube st=bst=b, insert the radial point with radius ∣s(a)∣|s(a)|. Its reduction on the new component has nonzero finite scaled coordinate, so is smooth. Points already specializing to smooth points are unchanged by a blowup supported at a node. Repeating at most ∣P∣|P| times puts every marked point in a smooth tube.

Suppose a smooth tube contains distinct marked points with coordinates a1,…,ama_1,\ldots,a_m, where m≥2m\geq 2. Set R=max⁡i,j∣ai−aj∣∈∣K×∣,0<R<1.R=\max_{i,j}|a_i-a_j|\in|K^\times|,\qquad 0<R<1. Insert ζa1,R\zeta_{a_1,R}. Every marked point lies in the closed disc ∣z−a1∣≤R|z-a_1|\leq R, so none lies in the new attaching annulus. On the exceptional affine line their specializations are the residues of (ai−a1)/λ(a_i-a_1)/\lambda, where ∣λ∣=R|\lambda|=R. At least two of these residues differ: one is zero and another has absolute value one. The points are therefore partitioned into at least two nonempty groups, each lying in a distinct smooth residual tube. Repeat on any group of size greater than one. Each step increases the number of nonempty groups and never exceeds ∣P∣|P|, so the process terminates with distinct smooth specializations. All preceding blowups retain old vertices; insert any additional prescribed type II vertices first. ◻

4.28 Smooth tubes with arbitrary residue extensions

Lemma 4.28.1. NAG-002094 (Scalar extension of algebraic formal models). Let L/KL/K be a complete isometric valued extension and let U\mathcal U be a flat finitely presented OO-scheme. Put O′=L∘O'=L^\circ, l=L~l=\widetilde L, and U′=U×OO′\mathcal U'=\mathcal U\times_O O'. Completion uses the same π∈O\pi\in O, 0<∣π∣<10<|\pi|<1, on both sides. There are canonical identifications (U′^η)≃(U^η)L,Us′=Us×kl,(\widehat{\mathcal U'}_\eta)\simeq (\widehat{\mathcal U}_\eta)_L, \qquad \mathcal U'_s=\mathcal U_s\times_k l, compatible with specialization, formal-open restriction, and successive scalar extensions. In particular a geometrically reduced special fibre remains reduced after this operation.

Proof. Work first with B=O[T1,…,Tn]/(f1,…,fs)B=O[T_1,\ldots,T_n]/(f_1,\ldots,f_s). Base change preserves flatness and finite presentation. Lemma 4.13.1, applied over OO and over O′O', gives generic algebras A=K⟨T⟩/(f1,…,fs),A′=L⟨T⟩/(f1,…,fs).A=K\langle T\rangle/(f_1,\ldots,f_s),\qquad A'=L\langle T\rangle/(f_1,\ldots,f_s). Proposition 2.33.5 identifies A′A' with A⊗^KLA\widehat\otimes_K L, with the actual extended relation ideal. Reduction of a restricted series modulo the maximal ideal leaves a polynomial: its coefficients tend to zero, so only finitely many have nonzero reduction. Conversely a restricted series with zero reduction has a largest coefficient absolute value less than one and belongs to the maximal ideal times the restricted series ring. Thus the special-fibre quotients are k[T]/(fˉi)k[T]/(\bar f_i) and l[T]/(fˉi)l[T]/(\bar f_i), which gives the second identification.

Principal affine restrictions have the common relation bU−1bU-1 by Lemma 4.13.1; their generic domains are ∣b∣=1|b|=1 on both sides. The local identifications therefore agree on restrictions and glue, using Theorem 2.40.12. Specialization commutes because its prime is defined by the integral functions with value less than one, and those values are unchanged by the isometric coefficient map. All comparisons agree on polynomial coordinates, so their composites for successive extensions agree as well. The last assertion is the definition of geometric reducedness. ◻

Lemma 4.28.2. NAG-002095 (Lifting an arbitrary finite residue extension). For every finite extension l/kl/k, there is a finite valued extension F/KF/K with F~≃l\widetilde F\simeq l over kk and [F:K]=[l:k],∣F×∣=∣K×∣.[F:K]=[l:k],\qquad |F^\times|=|K^\times|. If char⁡K=0\operatorname{char}K=0, the extension F/KF/K is separable.

Proof. For a simple extension l=k(β)l=k(\beta), use the construction in Lemma 4.15.2: lift the monic irreducible polynomial of β\beta, adjoin a root, and apply the value-group and residue-degree inequality. Every step through the displayed chain of degree inequalities in that proof uses only irreducibility, so it gives the asserted residue field, degree, and value group for this simple extension. A general finite extension has a finite tower of simple extensions. Apply the construction successively, using completeness and the unique extending absolute value from Lemma 1.12.1 at every step. Degrees multiply and the value group does not change. Finally, in characteristic zero an irreducible polynomial has nonzero derivative relatively prime to it, so all finite extensions are separable. ◻

Lemma 4.28.3. NAG-002096 (Galois rationalization of a residue-point support). Assume char⁡K=0\operatorname{char}K=0. For every finite extension κ/k\kappa/k, there exists a finite Galois extension N/KN/K such that all points of Spec⁡(κ⊗kN~)\operatorname{Spec}(\kappa\otimes_k\widetilde N) are N~\widetilde N-rational and Gal⁡(N/K)\operatorname{Gal}(N/K) acts transitively on this finite set of points.

Proof. By [StacksAlg] (Tag 030K), write κ/κs/k\kappa/\kappa_s/k with κs/k\kappa_s/k separable and κ/κs\kappa/\kappa_s purely inseparable. Choose a finite Galois extension ls/kl_s/k containing the normal closure of κs/k\kappa_s/k. There is a finite purely inseparable extension l/lsl/l_s containing every conjugate of κ\kappa over kk: for finitely many generators βi\beta_i of κ\kappa, choose powers βipei∈κs\beta_i^{p^{e_i}}\in\kappa_s, and adjoin to lsl_s the unique peip^{e_i}-th roots of their images under all embeddings κs↪ls\kappa_s\hookrightarrow l_s. When char⁡k=0\operatorname{char}k=0, take l=lsl=l_s. The field ll therefore contains the image of every kk-embedding of κ\kappa into an algebraic closure.

Lift ls/kl_s/k to a finite Galois valued extension E/KE/K by Lemma 4.18.1. Over EE, apply Lemma 4.28.2 to l/lsl/l_s, obtaining F/EF/E with residue field ll. Since KK has characteristic zero, F/KF/K is separable. Adjoining the finitely many roots of minimal polynomials of generators of F/KF/K gives a finite normal extension N/KN/K containing FF, by [StacksAlg] (Tag 09HU). It is Galois, and N~\widetilde N contains ll.

Any residue field of the finite N~\widetilde N-algebra κ⊗kN~\kappa\otimes_k\widetilde N embeds in an algebraic closure of N~\widetilde N. Its image is generated by N~\widetilde N and a kk-embedded copy of κ\kappa. The latter is already in ll, so that residue field is N~\widetilde N. Thus the points are precisely the distinct kk-embeddings of κ\kappa into N~\widetilde N. Restriction from Gal⁡(N/K)\operatorname{Gal}(N/K) onto Gal⁡(E/K)\operatorname{Gal}(E/K) is surjective by [StacksAlg] (Tag 0BME); the latter group induces Gal⁡(ls/k)\operatorname{Gal}(l_s/k). It acts transitively on the embeddings of κs\kappa_s, and a purely inseparable extension has at most one extension of any embedding. All embeddings of κ\kappa exist in ll, so this proves transitivity. ◻

Theorem 4.28.4. NAG-002097 (Canonical smooth-tube contraction in characteristic zero). Assume char⁡K=0\operatorname{char}K=0, and let qq be any closed point where a curve model X/O\mathcal X/O is smooth of relative dimension one. Let η\eta be the unique component through qq. Then Cq=]q[‾=]q[∪{vη}C_q=\overline{]q[}=]q[\cup\{v_\eta\} has a canonical jointly continuous strong deformation retraction onto vηv_\eta, functorial for model isomorphisms and isometric ground-field automorphisms. The tube is path connected and has the canonical Gauss-ray germ and path-connected collar neighborhood basis at vηv_\eta described in Proposition 4.25.2.

Proof. Choose an affine neighborhood U\mathcal U of qq which is smooth of relative dimension one over OO. Its special fibre is geometrically reduced by [StacksAlg] (Tag 056T); smoothness also gives a unique component through qq. Apply Lemma 4.28.3 to κ=κ(q)\kappa=\kappa(q). Lemma 4.28.1 identifies the formal generic fibre of UN∘\mathcal U_{N^\circ} with UNU_N and its reduced smooth special fibre with (Us)N~(\mathcal U_s)_{\widetilde N}. The finitely many points qjq_j above qq are rational smooth points. Thus their tubes DjD_j are open unit discs and their compact closures are Cj=Dj∪{vj}C_j=D_j\cup\{v_j\} by Lemma 4.21.1.

The projection p:UN→Up:U_N\to U is the open Galois quotient of Theorem 4.9.2. Specialization gives p−1(Dq)=∐jDjp^{-1}(D_q)=\coprod_jD_j. Flat going down, [StacksAlg] (Tag 00HS), maps all vjv_j to vηv_\eta. Lemma 4.20.1 therefore gives the displayed closure, and openness gives p−1(Cq)=⋃jCj.p^{-1}(C_q)=\bigcup_j C_j. The finite closed pasting and compact-product quotient argument in Proposition 4.22.2 now applies verbatim to the canonical disc homotopies: its topological proof requires exactly this equality, rational disc tubes upstairs, and full semilinear coordinate invariance. These have just been verified. It gives the desired jointly continuous contraction.

Two choices of NN embed over KK into a common finite Galois extension, obtained by taking a splitting field of their defining polynomials. Under further extension, a rational residual point remains a single rational point, and its open-disc tube is its analytic scalar extension by Lemma 4.28.1. The full ground-field compatibility in Theorem 4.21.2 makes the two constructions identical upstairs; surjectivity makes their descents equal. Shrinking U\mathcal U does not change the whole tube or its component endpoint, and the same comparison proves independence of U\mathcal U, functoriality, and semilinear invariance.

Finally, the Galois group acts transitively on the qjq_j by Lemma 4.28.3. All the collar and ray arguments of Proposition 4.25.2 use only this transitivity, finitely many disc-coordinate transitions, and the topological quotient. They therefore give path connectedness, the cofinal connected collars, and the canonical embedded ray with parameter −log⁡r-\log r in the present setting as well. ◻

Corollary 4.28.5. NAG-002098 (Global skeletons in characteristic zero). Assume char⁡K=0\operatorname{char}K=0. In Theorem 4.23.3, the hypothesis that the smooth closed points have separable residue fields can be omitted. The same conclusion holds for the branch-star correspondence of Theorem 4.25.3. Equivariant homotopies of such models descend under the finite Galois hypotheses of Corollary 4.23.4.

Proof. The proof of Proposition 4.23.2 uses separability only to obtain the completed smooth-tube contractions. Replace that input by Theorem 4.28.4. The cofinite whole-tube estimate is unchanged, so its proof gives the same compact stars and uniform continuity in time. Finite closed pasting then proves the global skeleton theorem with the same nodal intervals and edge lengths. For branch stars, replace the connected collar input by the same theorem; every topological argument and every normalization label is unchanged. Model functoriality makes the homotopy equivariant, so Corollary 4.9.3 descends it. The finite-graph quotient argument is purely topological and again applies without change. ◻

4.29 Intrinsic residual curves and metric directions

For a point xx admitting the cofinal branch stars of Theorem 4.25.3, define its set of outgoing directions by TxX=lim←⁡U∋xπ0(U∖{x}),T_xX=\varprojlim_{U\ni x}\pi_0(U\setminus\{x\}), where UU ranges through ordinary open neighborhoods, ordered by shrinking. The cofinal stars make this inverse system constant on their residual-point labels. In particular this definition depends only on the analytic topology.

Lemma 4.29.1. NAG-002128 (Comparison of annular collars at one direction). Suppose two nodal models of the same analytic curve satisfy Theorem 4.25.3, have xx as a component vertex, and d∈TxXd\in T_xX corresponds in each model to a rational point of the normalized residual component. Their outgoing Gauss-ray germs and parameters ℓ=−log⁡r\ell=-\log r coincide. More precisely, if z,wz,w are their outward-normalized collar coordinates, there is r0<1r_0<1 such that the two collars of inner radius r>r0r>r_0 are the same analytic domain and ∣z∣=∣w∣|z|=|w| there.

Proof. Write Az(a)={a<∣z∣<1}A_z(a)=\{a<|z|<1\} and similarly for ww. The cofinality in Theorem 4.25.3 allows alternating refinements of the two stars. Since both branches represent dd, we obtain numbers a0,a1,b0<1a_0,a_1,b_0<1 with Aw(b0)⊂Az(a0),Az(a1)⊂Aw(b0).A_w(b_0)\subset A_z(a_0),\qquad A_z(a_1)\subset A_w(b_0). Moreover, sufficiently thin collars of either model lie in any prescribed collar of the other. To see this directly, make that collar the selected component of a smaller star, and refine this star using the other model.

The analytic function zz is invertible on Aw(b0)A_w(b_0). Corollary 4.10.3 gives ∣z(y)∣=c ∣w(y)∣n(y∈Aw(b0)),c>0,n∈Z.|z(y)|=c\,|w(y)|^n \quad(y\in A_w(b_0)),\qquad c>0,\quad n\in\mathbf Z. As the ww-Gauss radius tends to 11, the point approaches xx through every sufficiently thin zz-collar. Hence ∣z∣→1|z|\to 1, so c=1c=1. Since both coordinate absolute values lie strictly between zero and one, n>0n>0. Applying the same argument to ww on Az(a1)A_z(a_1) gives ∣w∣=∣z∣m|w|=|z|^m with m>0m>0. On this nonempty collar, composition gives ∣z∣=∣z∣mn|z|=|z|^{mn}; therefore mn=1mn=1 and m=n=1m=n=1.

Choose r0>max⁡(a0,a1,b0)r_0>\max(a_0,a_1,b_0). For r>r0r>r_0 the first inclusion and ∣z∣=∣w∣|z|=|w| give Aw(r)⊂Az(r)A_w(r)\subset A_z(r); the second gives the converse. The analytic-domain universal property identifies their analytic structures. The identity between these two standard annulus presentations is now an analytic isomorphism preserving radius. Corollary 4.19.1 identifies the Gauss points and their radial homotopies. In particular the ray germs and the parameter −log⁡r-\log r agree, with the original absolute value unchanged. ◻

Theorem 4.29.2. NAG-002129 (The canonical residual curve at a component vertex). Let KK be algebraically closed, complete and nontrivially valued. Suppose that xx is a component vertex of a proper nodal curve model of an analytic curve XX. There is a normal proper kk-curve CxC_x, canonically characterized by k(Cx)=H(x)~.k(C_x)=\widetilde{\mathcal H(x)}. For every such model, CxC_x is canonically the normalization of its component corresponding to xx. The bijection between TxXT_xX and the closed points of CxC_x, the outgoing Gauss-ray germ in each direction, and its parameter ℓ=−log⁡r\ell=-\log r are independent of the model.

If ff is an analytic germ at xx with f(x)≠0f(x)\ne 0, choose c∈K×c\in K^\times with ∣cf(x)∣=1|cf(x)|=1 and put φ=cf~x∈k(Cx)×\varphi=\widetilde{cf}_x\in k(C_x)^\times. Then for the direction corresponding to p∈Cxp\in C_x one has, for all sufficiently small ℓ≥0\ell\geq 0, −log⁡∣f(γp(ℓ))∣=−log⁡∣f(x)∣+ord⁡p(φ)ℓ.-\log|f(\gamma_p(\ell))| =-\log|f(x)|+\operatorname{ord}_p(\varphi)\ell . Only finitely many of these slopes are nonzero, and their sum is zero.

Proof. Let CC be the special component corresponding to xx. Lemma 4.24.2 identifies its function field with H(x)~\widetilde{\mathcal H(x)} and gives ∣H(x)×∣=∣K×∣|\mathcal H(x)^\times|=|K^\times|. The normalization CνC^\nu is finite over CC by [StacksAlg] (Tag 0BXR), hence is proper by [StacksAlg] (Tags 01WM, 01W3). Here CC is a closed component of the proper special fibre; properness is preserved by base change and closed immersion [StacksAlg] (Tags 01W4, 01W5). The equivalence between normal proper curves and their function fields [StacksAlg] (Tag 0BY1) identifies any two such normalizations by the unique isomorphism inducing the identity of H(x)~\widetilde{\mathcal H(x)}. This defines CxC_x.

The direction set itself is intrinsic by the cofinal description preceding the lemma. Every closed residual point is rational, since kk is algebraically closed. Lemma 4.29.1 therefore identifies the ray germs and their parameters for the same intrinsic direction in any two models. It remains to check that the associated residual points correspond under the just constructed isomorphism of normal proper curves.

Every ψ∈H(x)~×\psi\in\widetilde{\mathcal H(x)}^\times is the residue of a norm-one analytic unit germ at xx. Indeed, the proof of Lemma 4.24.2 represents it by a fraction P/QP/Q of affinoid functions with P(x)Q(x)≠0P(x)Q(x)\ne 0; on a small neighborhood QQ, and also PP, are invertible by Proposition 2.2.4. Apply Theorem 4.24.5 to this germ in the two models. The rays and length parameters coincide, so the two residual-point valuations take the same value on every ψ\psi. Thus their discrete valuation rings in the common function field coincide. They have the same centre on the identified proper curve: the two maps from the spectrum of this valuation ring agree at its generic point, and the closed diagonal of the separated curve makes them equal. Hence the residual points coincide.

The asserted formula is now Theorem 4.24.5 in any one model. Finite support and the sum-zero assertion are Theorem 4.24.7; here all closed-point degrees are one. Multiplying the normalizing constant cc by a norm-one scalar multiplies φ\varphi by a nonzero constant and does not change any order. ◻

Corollary 4.29.3. NAG-002130 (Principal slopes at arbitrary type II points). Let KK be algebraically closed, complete and nontrivially valued. Let YY be a smooth integral projective KK-curve possessing a projective nodal OO-model, and put X=YanX=Y^{\mathrm{an}}. Every type II point x∈Xx\in X has the intrinsic residual curve, metric directions and local slope formula of Theorem 4.29.2. For every f∈K(Y)×f\in K(Y)^\times, there is a finite marked skeleton of X∖∣div⁡(f)∣X\setminus|\operatorname{div}(f)| such that −log⁡∣f∣-\log|f| factors through its retraction, is integral affine on every open edge and ray, and has outgoing sum zero at every finite vertex. On a ray ending at a∈∣div⁡(f)∣a\in|\operatorname{div}(f)| its slope towards aa is ord⁡a(f)\operatorname{ord}_a(f). Any prescribed finite set of type II points can be included among the vertices.

Proof. The generic-fibre comparison of Proposition 4.13.2 identifies the formal generic fibre with XX. Theorem 4.27.4 realizes any chosen type II point, or any finite set of them, as component vertices of a projective nodal model. Theorem 4.29.2 applies at each such vertex and proves that the resulting objects do not depend on that choice.

The support P=∣div⁡(f)∣P=|\operatorname{div}(f)| is finite and consists of KK-rational points. Apply Theorem 4.27.5 while retaining the prescribed vertices. The marked skeleton and every stated norm, affine and slope identity then follow from Theorem 4.26.2 and Theorem 4.26.3. ◻

4.30 Models in genus zero and over a discrete valuation ring

Proposition 4.30.1. NAG-002121 (An explicit genus-zero model). Let KK be complete and nontrivially valued of characteristic zero, with the fixed absolute value, and let C/KC/K be a smooth projective geometrically connected curve of genus zero. There is a finite Galois extension L/KL/K and an isomorphism CL≃PL1C_L\simeq\mathbf P^1_L. The smooth projective model PL∘1\mathbf P^1_{L^\circ} gives a one-vertex skeleton and a jointly continuous strong deformation retraction of (CL)an(C_L)^{\mathrm{an}} onto that vertex. The value group of KK may be nondiscrete.

Proof. A nonempty affine open of CC has a closed point with finite residue field over KK, by [StacksAlg] (Tag 00FV). Take a finite normal closure of this field over KK. Characteristic zero makes the resulting L/KL/K Galois, and CLC_L has an LL-rational point PP. Smoothness makes PP an effective Cartier divisor by [StacksAlg] (Tag 0B8Y). Thus O(P)\mathcal O(P) has degree one. Flat cohomology base change [StacksAlg] (Tag 02KH) gives H1(CL,O)=H1(C,O)⊗KL=0H^1(C_L,\mathcal O)=H^1(C,\mathcal O)\otimes_K L=0. The characterization [StacksAlg] (Tag 0C6U) yields CL≃PL1C_L\simeq\mathbf P^1_L.

The two standard affine charts of PL∘1\mathbf P^1_{L^\circ} are finitely presented and flat; the special fibre is the smooth integral curve PL~1\mathbf P^1_{\widetilde L}. Its completion has analytic generic fibre (CL)an(C_L)^{\mathrm{an}} by Theorem 4.13.2. Corollary 4.28.5 applies with no nodal edges and one component vertex, and gives the asserted homotopy. ◻

For the next three results, RR is a complete discrete valuation ring with fraction field FF, uniformizer ϖ\varpi, and residue field ll. A split nodal model is a flat, proper, finitely presented relative curve with smooth generic fibre, whose special-fibre singularities are ll-rational split nodes. Its other special-fibre points are smooth. At each node we include a rational pointed étale chart over RR of the form supplied by Theorem 4.18.2. The word regular refers to its total space.

Lemma 4.30.2. NAG-002122 (Equivariant resolution of a split nodal model). Let Y/R\mathcal Y/R be a split nodal model. Repeatedly blow up the reduced union of all nonregular points of the total space. This process terminates with a regular split nodal model Y′/R\mathcal Y'/R. It preserves projectivity and is functorial for semilinear isomorphisms preserving RR. In particular every finite semilinear group action on Y\mathcal Y lifts to Y′\mathcal Y'.

Proof. Nonregular points lie among the finitely many special nodes. By the pointed charts of Theorem 4.18.2, after absorbing a unit into one coordinate their local equation is xy=ϖnxy=\varpi^n, n≥1n\geq 1, in an étale neighborhood with the same residue field. The completed local ring is R[[x,y]]/(xy−ϖn)R[[x,y]]/(xy-\varpi^n) by [StacksAlg] (Tag 0AGX); it is regular precisely for n=1n=1. For n≥2n\geq 2, blow up (x,y,ϖ)(x,y,\varpi). The ϖ\varpi-chart has equation x1y1=ϖn−2x_1y_1=\varpi^{n-2}, where x=ϖx1x=\varpi x_1 and y=ϖy1y=\varpi y_1. On the xx-chart put u=ϖ/xu=\varpi/x; after removing the cancelled torsion the equation is xu=ϖxu=\varpi, with y=xn−1uny=x^{n-1}u^n. The yy-chart is symmetric. These are the local charts computed in the proof of [StacksAlg] (Tag 0CDE); flat base change of blowups is [StacksAlg] (Tag 0805).

All new special nodes are rational and split. The two outer charts are regular. In the middle chart, n=2n=2 gives a smooth chart, n=3n=3 a regular node, and n≥4n\geq 4 gives a unique remaining nonregular point with exponent n−2n-2. Around each center choose an affine or étale neighborhood omitting the other finitely many centers. The other center ideals are units there, so blowup of the finite reduced union restricts to exactly the local calculation just made. Every round strictly decreases each surviving exponent; after finitely many rounds no nonregular point remains. The center ideal is of finite type on this Noetherian scheme, so blowups are projective by [StacksAlg] (Tag 02NS); composition preserves projectivity over the affine base by [StacksAlg] (Tag 0C4P). They leave the generic fibre unchanged. Finally the nonregular locus and its reduced ideal are invariant under every semilinear isomorphism. Such an isomorphism acts on all powers of that ideal, hence on its Rees algebra and its Proj. These induced maps commute with composition, giving the asserted functoriality. ◻

Lemma 4.30.3. NAG-002123 (Separating marked sections). Let Y/R\mathcal Y/R be a regular split nodal model and let P1,…,PmP_1,\ldots,P_m be distinct FF-rational points of its generic fibre. Their closures are sections meeting the smooth locus. Blowing up, at each stage, the reduced union of points at which at least two marked sections meet terminates in a regular split nodal model on which the marked sections are disjoint. Projectivity is preserved. If a finite semilinear group acts on the model and permutes the marked generic points, it acts on every stage and on the final marked model.

Proof. Properness extends each PiP_i uniquely to a section by [StacksAlg] (Tag 0BX5). Its special point is ll-rational. At a split node the regular completed local equation is xy=ϵϖxy=\epsilon\varpi, with ϵ\epsilon a unit, by the preceding local computation. A section specializing to the node would give x,y∈ϖRx,y\in\varpi R, contradicting xy=ϵϖxy=\epsilon\varpi. One may evaluate this completed equation because a section through the point induces a continuous local map to the complete ring RR. Consequently every marked section meets the smooth locus.

At a collision point choose a smooth étale coordinate TT reducing to zero. The completed local ring is R[[T]]R[[T]], by the smooth pointed chart and [StacksAlg] (Tag 0AGX). A section through this point has T=ai∈ϖRT=a_i\in\varpi R. Distinct sections have distinct aia_i: the rational smooth-tube isomorphism of Corollary 4.17.2 is injective on their generic analytic points. For two sections with the same special point set cij=ord⁡ϖ(ai−aj)c_{ij}=\operatorname{ord}_{\varpi}(a_i-a_j), and set cij=0c_{ij}=0 if their special points differ. This nonnegative integer is coordinate independent. Indeed a change of smooth coordinate in R[[T]]R[[T]] has the form g(T)=b0+b1T+⋯g(T)=b_0+b_1T+\cdots, with b0∈ϖRb_0\in\varpi R, b1∈R×b_1\in R^\times, and bn∈Rb_n\in R. For ai,aj∈ϖRa_i,a_j\in\varpi R the quotient (g(ai)−g(aj))/(ai−aj)(g(a_i)-g(a_j))/(a_i-a_j) is a unit.

Blow up the collision point, whose ideal is (ϖ,T)(\varpi,T). By [StacksAlg] (Tag 0805) its two local charts are the pullbacks of Spec⁡R[U],T=ϖU;Spec⁡R[T,V]/(TV−ϖ).\operatorname{Spec}R[U],\quad T=\varpi U; \qquad \operatorname{Spec}R[T,V]/(TV-\varpi). They are regular, and their special fibres are smooth or split nodal. Every marked section through the center lifts to the first chart, where U=ai/ϖU=a_i/\varpi. Existence and uniqueness also follow from the blowup universal property [StacksAlg] (Tag 0806), since the pulled-back center ideal is (ϖ)(\varpi). Thus marked sections remain smooth. For every pair meeting at the center, the new contact number is exactly cij−1c_{ij}-1; this includes cij=1c_{ij}=1, when their new special points are distinct. All other contact numbers are unchanged. The finite sum ∑i<jcij\sum_{i<j}c_{ij} strictly decreases whenever a collision remains. It proves termination with disjoint sections.

The collision locus is a finite set of rational smooth points, invariant under any automorphism permuting the marked sections. Blowup of its reduced union therefore lifts the group action by the Rees-algebra argument in Lemma 4.30.2. Projectivity follows again from [StacksAlg] (Tags 02NS, 0C4P). The generic marked points remain unchanged throughout, and uniqueness of their extensions keeps their labels and their permutation action compatible. ◻

Theorem 4.30.4. NAG-002124 (Split marked models over a complete discrete field). Let KK be a complete nontrivially discretely valued field of characteristic zero. Let C/KC/K be a smooth projective geometrically connected curve, and let DD be a finite reduced closed subscheme of CC. There is a finite Galois extension L/KL/K such that DLD_L is a disjoint union of LL-rational points and CLC_L has a regular projective split nodal L∘L^\circ-model on which their sections are disjoint and smooth. If g(C)>0g(C)>0, the model may be chosen with a semilinear Gal⁡(L/K)\operatorname{Gal}(L/K)-action extending the given action on CLC_L and permuting these sections. All absolute values extend the specified one on KK.

Proof. After extension to an algebraic closure, smoothness makes all local rings regular, so distinct irreducible components cannot meet. Geometric connectedness therefore makes CC geometrically integral. By [StacksAlg] (Tag 04L2(4)), H0(C,OC)=KH^0(C,\mathcal O_C)=K. Apply the all-genus semistable reduction theorem [StacksAlg] (Tag 0CDN). Its extension of discrete valuation rings is the ring of integers for the unique extending absolute value: if ϖ0\varpi_0 is a uniformizer of K∘K^\circ and its order in the extension ring is e>0e>0, the formula ∣b∣′=∣ϖ0∣ord⁡(b)/e|b|'=|\varpi_0|^{\operatorname{ord}(b)/e} extends the given absolute value. Uniqueness in Lemma 1.12.1 identifies it with the fixed extension. Finite extension preserves a cyclic nontrivial value group by Lemma 3.24.2.

Enlarge this field to a finite Galois extension L0/KL_0/K also splitting all residue fields of DD. These fields are separable since KK has characteristic zero. The base-changed semistable model remains nodal. Choose a minimal regular proper model M\mathcal M over L0∘L_0^\circ, using [StacksAlg] (Tag 0C2W). It is nodal by [StacksAlg] (Tag 0CDG), and it is projective by [StacksAlg] (Tag 0C5P), applied to its regular two-dimensional total space over the Noetherian base Spec⁡L0∘\operatorname{Spec}L_0^\circ.

Use Theorem 4.18.2 on the finitely many nodes of Ms\mathcal M_s, and take a finite Galois normal closure L/KL/K of the resulting extension. Split rational node equations remain split rational node equations after this further base change. Thus M×L0∘L∘\mathcal M\times_{L_0^\circ}L^\circ is a projective split nodal model. Apply Lemma 4.30.2 and then Lemma 4.30.3. The result has all the claimed geometric properties.

Suppose g(C)>0g(C)>0. The unique minimal model theorem [StacksAlg] (Tag 0C6B) extends the generic semilinear Gal⁡(L0/K)\operatorname{Gal}(L_0/K)-action to M\mathcal M: apply uniqueness to M\mathcal M and each coefficient twist with the prescribed generic identification. The extensions are unique, so their compositions satisfy the group law. Since L0/KL_0/K and L/KL/K are Galois, this gives a semilinear action of Gal⁡(L/K)\operatorname{Gal}(L/K) on M×L0∘L∘\mathcal M\times_{L_0^\circ}L^\circ by restriction on the first factor and its natural action on the second. Both subsequent constructions are equivariant by the two preceding lemmas. ◻

4.31 Geometric slopes over an imperfect residue field

Lemma 4.31.1. NAG-002131 (Smooth normalized residual components). Let DD be a reduced finite-type curve over a field kk. Assume that DD is smooth away from finitely many split kk-rational nodes. Then its normalization DνD^\nu is smooth over kk. For every field extension l/kl/k, DlD_l is reduced and the finite morphism (Dν)l⟶Dl(D^\nu)_l\longrightarrow D_l is the normalization. The same statements hold for each irreducible component CC and its normalization CνC^\nu.

Proof. Normalization is finite by [StacksAlg] (Tag 0BXR). Away from the nodes it is an isomorphism, because the smooth local rings are normal. At a node, Lemma 4.24.1 identifies the two normalization points with the origins of the two axes through actual Cartesian normalization squares for the pointed étale charts. These points are rational, and their local rings on the normalized components are discrete valuation rings. By [StacksAlg] (Tag 0B8Y), a normal curve is smooth at a closed point whose residue field is separable over kk. Thus DνD^\nu is smooth at these points as well. This proves smoothness everywhere, also when the two node branches belong to the same component.

Consequently (Dν)l(D^\nu)_l is normal and reduced for every l/kl/k, by [StacksAlg] (Tag 056T). On an affine open of DD, write its finite normalization as R↪SR\hookrightarrow S. The inclusion is injective because RR is reduced. Tensoring with the flat field extension l/kl/k gives R⊗kl↪S⊗klR\otimes_k l\hookrightarrow S\otimes_k l. The right side is reduced, so DlD_l is reduced. This argument applies separately to each CC as well, since CνC^\nu is a component of the smooth scheme DνD^\nu.

The base-changed finite map is an isomorphism off the inverse image of the finite node set. This open set contains every generic point of DlD_l: flat going down [StacksAlg] (Tag 00HS) sends each minimal prime to a minimal prime of DD. The same holds for ClC_l. Thus on each affine chart the two rings have the same total ring of fractions. The source ring is normal and integral over the target ring. An element of their common total fraction ring integral over the target is also integral over the source, hence belongs to the source; the reverse inclusion follows from finiteness. Therefore the source is exactly the integral closure, proving the asserted normalization comparison. ◻

Lemma 4.31.2. NAG-002132 (Residual tensor lengths and orders). Let CC be a smooth integral curve over an arbitrary field kk, let pp be a closed point, and write κ=κ(p),s=[κ:k]sep,e=[κ:k]insep.\kappa=\kappa(p),\qquad s=[\kappa:k]_{\mathrm{sep}},\qquad e=[\kappa:k]_{\mathrm{insep}}. Choose a field extension l/kl/k containing the images of all kk-embeddings of κ\kappa into an algebraic closure of kk. Then plp_l has exactly ss distinct support points, all ll-rational, and its local ring at each support point has length ee. If p′p' is any such point of the smooth curve ClC_l, then for every φ∈k(C)×\varphi\in k(C)^\times, ord⁡p′(φ)=e ord⁡p(φ).\operatorname{ord}_{p'}(\varphi) =e\,\operatorname{ord}_p(\varphi). Here φ\varphi is pulled back to the component of ClC_l through p′p'.

Proof. Let κs\kappa_s be the maximal separable subextension; then κ/κs\kappa/\kappa_s is purely inseparable by [StacksAlg] (Tag 030K). Since ll contains all the embeddings of κs\kappa_s, the usual factorization of a separable minimal polynomial gives κ⊗kl=∏σ:κs↪l(κ⊗κs,σl).\kappa\otimes_k l =\prod_{\sigma:\kappa_s\hookrightarrow l} \bigl(\kappa\otimes_{\kappa_s,\sigma}l\bigr). Each factor has dimension ee over ll. Let σ^:κ↪l\widehat\sigma:\kappa\hookrightarrow l be the unique extension of σ\sigma. It gives a surjection from that factor to ll. In positive characteristic, choose finitely many generators βi\beta_i of κ/κs\kappa/\kappa_s and powers βipni∈κs\beta_i^{p^{n_i}}\in\kappa_s. The kernel is generated by βi⊗1−1⊗σ^(βi)\beta_i\otimes 1-1\otimes\widehat\sigma(\beta_i), each of which is nilpotent. Thus this is the unique maximal ideal. In characteristic zero each factor is simply ll. Every composition factor of the resulting Artin local ring is ll, so its length is its ll-dimension ee. There are precisely ss factors.

Put R=OC,pR=\mathcal O_{C,p} and R′=OCl,p′R'=\mathcal O_{C_l,p'}. Both rings are discrete valuation rings, since the two curves are smooth. If uu is a uniformizer of RR, then R′/uR′=(κ⊗kl)p′,ord⁡R′(u)=length⁡R′(R′/uR′)=e.R'/uR'=(\kappa\otimes_k l)_{p'},\qquad \operatorname{ord}_{R'}(u) =\operatorname{length}_{R'}(R'/uR')=e. The local ring map R→R′R\to R' carries units to units and is injective by flatness. Writing φ=umv\varphi=u^m v with v∈R×v\in R^\times therefore gives ord⁡p′(φ)=em\operatorname{ord}_{p'}(\varphi)=em for every m∈Zm\in\mathbf Z, as asserted. ◻

Theorem 4.31.3. NAG-002133 (Geometric slope formula in characteristic zero). Let KK be complete, nontrivially valued and of characteristic zero, and retain its original absolute value. Let X/O\mathcal X/O be a curve model satisfying Corollary 4.28.5. Let xx be the component vertex associated with CC. For a closed point p∈Cνp\in C^\nu, put ep=[κ(p):k]insep,sp=[κ(p):k]sep.e_p=[\kappa(p):k]_{\mathrm{insep}},\qquad s_p=[\kappa(p):k]_{\mathrm{sep}}. Use the geometric Gauss-ray germ γp\gamma_p supplied by that corollary, parameterized by ℓ=−log⁡r\ell=-\log r after a finite valued extension making the residual point rational. If ff is an analytic germ at xx with f(x)≠0f(x)\ne 0, choose c∈K×c\in K^\times such that ∣cf(x)∣=1|cf(x)|=1, and let φ=cf~x∈k(C)×=k(Cν)×.\varphi=\widetilde{cf}_x \in k(C)^\times=k(C^\nu)^\times. Then for all sufficiently small ℓ≥0\ell\geq 0, −log⁡∣f(γp(ℓ))∣=−log⁡∣f(x)∣+ep ord⁡p(φ)ℓ.-\log|f(\gamma_p(\ell))| =-\log|f(x)|+ e_p\,\operatorname{ord}_p(\varphi)\ell . The formula and its parameter are independent of the chosen finite valued extension. In particular the geometric outgoing slope is the integer σp(f)=ep ord⁡p(φ).\sigma_p(f) =e_p\,\operatorname{ord}_p(\varphi).

Proof. Lemma 4.24.2 gives the normalizing scalar cc and the displayed residual function field. If pp lies over a node, it is rational by Lemma 4.24.1; hence ep=1e_p=1 and the assertion is Theorem 4.24.5 in its actual pointed node chart.

Otherwise pp corresponds to a smooth point qq of the special fibre, with κ(p)=κ(q)\kappa(p)=\kappa(q). Choose the finite Galois extension N/KN/K in Lemma 4.28.3, with l=N~l=\widetilde N containing all the residue embeddings. By Lemma 4.28.1 the algebraic model base change has special fibre (Xs)l(\mathcal X_s)_l, and its formal generic fibre is the analytic scalar extension. Lemma 4.31.1 shows that this special fibre is reduced and that its normalization is (Xsν)l(\mathcal X_s^\nu)_l.

Choose a rational point p′p' above pp, and let x′x' be the component endpoint of its rational smooth tube. It maps to xx, so ∣cf(x′)∣=1|cf(x')|=1. The component-residue identifications commute with pullback: on integral affine functions both are ordinary special-fibre reduction, and the fraction-field identifications extend these maps. Thus the residue of cfcf at x′x' is the pullback of φ\varphi to the component through p′p'. Theorem 4.24.5 applied upstairs gives a geometric slope ord⁡p′(φ)\operatorname{ord}_{p'}(\varphi). By Lemma 4.31.2, this equals epord⁡p(φ)e_p\operatorname{ord}_p(\varphi).

Theorem 4.28.4 identifies the projection of the upstairs Gauss-ray germ with γp\gamma_p, using the same radius rr and the same extending absolute value. Evaluation of a function pulled back from XX is unchanged by projection. The upstairs equality therefore gives exactly the displayed equality on XX. The independence of extension and coordinate follows from the ray-germ comparison in that theorem. Multiplying cc by a norm-one scalar changes φ\varphi by a nonzero constant and leaves all orders unchanged. ◻

Corollary 4.31.4. NAG-002134 (Separable-degree balance). Under the hypotheses of Theorem 4.31.3, assume that CC is proper over kk. Only finitely many geometric slopes σp(f)\sigma_p(f) are nonzero, and ∑p∈Cν closedsp σp(f)=0.\sum_{p\in C^\nu\ {\rm closed}} s_p\,\sigma_p(f)=0. Equivalently, the algebraically normalized slopes σp(f)/ep\sigma_p(f)/e_p satisfy the residue-degree identity ∑p∈Cν closed[κ(p):k] σp(f)ep=0.\sum_{p\in C^\nu\ {\rm closed}} [\kappa(p):k]\,\frac{\sigma_p(f)}{e_p}=0. When kk is perfect, ep=1e_p=1 and this is Theorem 4.24.7.

Proof. The finite normalization CνC^\nu is proper by [StacksAlg] (Tags 0BXR, 01WM, 01W3). The principal divisor of φ\varphi has finite support. Its degree is zero by [StacksAlg] (Tag 02RU), so 0=∑p[κ(p):k]ord⁡p(φ)=∑pspepord⁡p(φ)=∑pspσp(f).0=\sum_p[\kappa(p):k]\operatorname{ord}_p(\varphi) =\sum_p s_p e_p\operatorname{ord}_p(\varphi) =\sum_p s_p\sigma_p(f). Both displayed assertions and finite support follow. ◻

4.32 Marked tubes and punctured skeleta

Lemma 4.32.1. NAG-002125 (The punctured-disc homotopy). Let DD be an open unit disc, let P∈D(K)P\in D(K), and adjoin its outer endpoint vv as in Lemma 4.21.1. Choose a disc coordinate TT with T(P)=0T(P)=0. On DP∗=(D∖{P})∪{v}D_P^*=(D\setminus\{P\})\cup\{v\} there is a canonical jointly continuous strong deformation retraction onto SP={v}∪{ζ0,r:0<r<1}.S_P=\{v\}\cup\{\zeta_{0,r}:0<r<1\}. The parameter ℓ=−log⁡r\ell=-\log r, with ℓ(v)=0\ell(v)=0, identifies SPS_P with the metric ray [0,∞)[0,\infty). The homotopy preserves ∣T∣|T|, is independent of the coordinate centered at PP, and commutes with complete isometric scalar extension and semilinear disc isomorphisms carrying the marked point to the marked point.

Proof. Using Theorem 4.21.2, define HP(x,u)=HD(x,u∣T(x)∣)(x≠v),HP(v,u)=v.H_P(x,u)=H_D\bigl(x,u|T(x)|\bigr) \quad(x\ne v),\qquad H_P(v,u)=v. The parameter on the right is less than one. The weighted Gauss formula gives ∣T(HP(x,u))∣=∣T(x)∣>0|T(H_P(x,u))|=|T(x)|>0. On D∖{P}D\setminus\{P\} joint continuity is composition of continuous maps. At vv the entire outer tail ∣T∣>r|T|>r is preserved for every uu, so the full-tail neighborhood basis gives joint continuity there, uniformly in uu. At u=1u=1 the radius-∣T(x)∣|T(x)| Gauss norm is unchanged by translation by T(x)T(x), giving HP(x,1)=ζ0,∣T(x)∣H_P(x,1)=\zeta_{0,|T(x)|}. The same formula fixes every centered Gauss point for every uu, as in Proposition 4.3.1. It starts at xx, and hence is the asserted strong retraction.

A centered change of coordinate has the form g(T)=a1T+a2T2+⋯g(T)=a_1T+a_2T^2+\cdots with ∣a1∣=1|a_1|=1 and ∣an∣≤1|a_n|\leq 1. Thus ∣g(x)∣=∣T(x)∣|g(x)|=|T(x)| throughout the punctured disc. The full coordinate invariance of HDH_D now proves that HPH_P is coordinate independent, not just its value at u=1u=1. The same argument proves semilinear compatibility and scalar-extension compatibility. Finally the map ℓ↦ζ0,exp⁡(−ℓ)\ell\mapsto\zeta_{0,\exp(-\ell)} is continuous for ℓ>0\ell>0 and extends to vv at zero by the tube endpoint theorem. Its inverse is the continuous function −log⁡∣T∣-\log|T|, extended by zero at vv. This proves the topological and metric ray identification. ◻

Theorem 4.32.2. NAG-002126 (A split marked model gives a punctured skeleton). Let C/O\mathcal C/O be a projective curve model satisfying either the hypotheses of Theorem 4.23.3 (finite separable residue fields at smooth closed points) or those of Corollary 4.28.5 (characteristic-zero KK, with arbitrary residue fields at smooth closed points). Let P1,…,PmP_1,\ldots,P_m be KK-rational points of its smooth generic fibre whose sections have distinct smooth special points q1,…,qmq_1,\ldots,q_m. Put U=(CK)an∖{P1,…,Pm}.U=(\mathcal C_K)^{\mathrm{an}}\setminus\{P_1,\ldots,P_m\}. Adjoin to the compact model skeleton one infinite ray for every PiP_i, contained in ]qi[]q_i[ and attached to its component vertex. Their union Σ(C;P1,…,Pm)⊂U\Sigma(\mathcal C;P_1,\ldots,P_m)\subset U is a finite metric graph with finitely many infinite rays and is a jointly continuous strong deformation retract of UU. The entire homotopy and all edge lengths are invariant under isomorphisms of marked models, including semilinear isometric isomorphisms.

Proof. The formal completion has generic fibre (CK)an(\mathcal C_K)^{\mathrm{an}} by Theorem 4.13.2. Each qiq_i is rational because it is the special point of an OO-section. Its tube is a rational open disc by Corollary 4.17.2. Translate a disc coordinate by its value at PiP_i to put that point at zero. Use Lemma 4.32.1 on the completed tube with PiP_i removed.

We justify the global pasting using a finite closed cover. For every component vertex vv, start with its compact star AvA_v from Proposition 4.23.2, using when necessary the arbitrary-residue extension of Corollary 4.28.5. Remove from AvA_v all the finitely many marked open tubes incident to it; the resulting Av0A_v^0 is closed in AvA_v and thus compact and closed in UU. The original star homotopy restricts to Av0A_v^0. Add the finitely many punctured completed marked tubes Cqi∖{Pi}C_{q_i}\setminus\{P_i\}, which are closed in UU, and the finitely many completed nodal tubes. These sets cover UU and their intersections consist of fixed vertices. Their local homotopies therefore paste to a jointly continuous map on U×[0,1]U\times[0,1] by the finite closed-cover pasting lemma. This argument retains the uniform control of all unmarked tubes already proved for each AvA_v.

At time one, unmarked stars map to their vertices, nodal tubes to their compact Gauss edges, and each marked punctured tube to its ray. Every point of this union is fixed at every time. It is therefore the strong retraction image. The graph topology is the one induced from UU: its compact core and each of its finitely many rays are closed subspaces, and they meet only at their attached endpoints. On a ray this follows from Lemma 4.32.1; on the compact core it is Theorem 4.23.3. For a finite closed cover these subspace topologies determine the topology on their union. The length coordinate on each ray is −log⁡∣T∣-\log|T|, and the compact edge lengths are the previously established annular lengths. The local homotopy invariances give all the asserted functoriality. ◻

Corollary 4.32.3. NAG-002127 (Punctured curves over the original discrete field). Let K,C,DK,C,D be as in Theorem 4.30.4. After its finite Galois extension L/KL/K, (CL∖DL)an(C_L\setminus D_L)^{\mathrm{an}} has a strong deformation retraction onto a finite metric graph with one infinite ray for each point of DLD_L. If g(C)>0g(C)>0, the homotopy descends to (C∖D)an(C\setminus D)^{\mathrm{an}} over the original field and the original absolute value. Its image is the finite metric quotient graph, with one infinite ray for each closed point of DD.

Proof. Apply Theorem 4.32.2 to the split marked projective model just constructed. For g(C)>0g(C)>0 that model has the compatible Galois action; the full marked homotopy is equivariant. Corollary 4.9.3, applied to the analytic open space (C∖D)an(C\setminus D)^{\mathrm{an}}, descends it. This descent is valid without compactness: Theorem 4.9.2 gives the open quotient property after product with [0,1][0,1].

The quotient of the compact core is a finite metric graph by Corollary 4.23.4. A group element carrying one marked disc to another carries its center to the other center. Hence it preserves the radius on the entire ray, by Lemma 4.32.1. Thus stabilizers act trivially on each ray, and quotient rays are indexed by Galois orbits of DLD_L, equivalently by the closed points of DD. All parameters are computed with the extending absolute value; no degree factor is introduced. ◻

4.33 Model-independent geometric directions in characteristic zero

Lemma 4.33.1. NAG-002153 (The norm profile on an entire residual collar). Assume Theorem 4.31.3. For every residual direction pp there are an open collar Bp(r0)B_p(r_0) and a continuous radius function Rp:Bp(r0)⟶(r0,1),Bp(r)={y∈Bp(r0):Rp(y)>r},R_p:B_p(r_0)\longrightarrow(r_0,1),\qquad B_p(r)=\{y\in B_p(r_0):R_p(y)>r\}, whose Gauss-ray representative has radius rr. The germ of this radius function is independent of the rationalizing extension and coordinates. For every analytic germ ff at the component vertex xx with f(x)≠0f(x)\ne 0, after increasing r0r_0 one has on all actual points of the collar ∣f(y)∣=∣f(x)∣ Rp(y) epord⁡p(cf~x),∣cf(x)∣=1.|f(y)|=|f(x)|\,R_p(y)^{\,e_p\operatorname{ord}_p (\widetilde{cf}_x)}, \qquad |cf(x)|=1 .

Proof. At a rational node branch use its standard annular coordinate. At a smooth residual point, the construction of Proposition 4.25.2, extended by Theorem 4.28.4, produces finitely many rational open discs after a finite Galois extension. Lemma 4.25.1 makes the finitely many semilinear transition maps preserve the entire radius functions on sufficiently thin collars. These functions therefore descend through the analytic Galois quotient to RpR_p. The same lemma and a common finite Galois extension compare two constructions. They also show that Bp(r)B_p(r) is exactly the image of any upstairs radius-rr outer collar.

By Proposition 2.2.4, ff and f−1f^{-1} are analytic on a neighborhood of xx. Cofinality of the branch stars puts a sufficiently thin collar inside that neighborhood. On every rational annular collar upstairs, Corollary 4.10.3 therefore gives an all-point identity ∣f∣=arn|f|=a r^n. Along its Gauss ray, Theorem 4.31.3 identifies a=∣f(x)∣a=|f(x)| and n=epord⁡p(cf~x)n=e_p\operatorname{ord}_p(\widetilde{cf}_x). These constants are the same on every conjugate collar. Evaluation and the radius function descend, proving the identity on the entire collar downstairs. ◻

Theorem 4.33.2. NAG-002154 (Intrinsic residual curve and geometric metric). Let KK be complete, nontrivially valued and of characteristic zero. Suppose that xx is a component vertex of a proper curve model satisfying Corollary 4.28.5. There is a canonical smooth proper residual curve CxC_x over kk with function field H(x)~\widetilde{\mathcal H(x)}. For every such model of the same analytic curve XX having xx as a component vertex, it identifies canonically with the normalization of the component corresponding to xx.

The correspondence between the intrinsic outgoing directions TxXT_xX and the closed points of CxC_x is independent of the model. So are the geometric radius germ RpR_p, the Gauss-ray germ and its parameter ℓ=−log⁡r\ell=-\log r in every direction. For every analytic germ ff nonzero at xx, the intrinsic geometric slopes are σp(f)=[κ(p):k]insep ord⁡p(cf~x),∣cf(x)∣=1,\sigma_p(f)=[\kappa(p):k]_{\mathrm{insep}}\, \operatorname{ord}_p(\widetilde{cf}_x), \qquad |cf(x)|=1 , and satisfy the finite sum ∑p∈Cx closed[κ(p):k]sep σp(f)=0.\sum_{p\in C_x\ {\rm closed}} [\kappa(p):k]_{\mathrm{sep}}\, \sigma_p(f)=0.

Proof. The component-residue identification Proposition 4.24.2 gives the function field. The finite normalization of a proper residual component is proper by [StacksAlg] (Tags 0BXR, 01WM, 01W3); Lemma 4.31.1 makes it smooth. The normal proper curve/function-field equivalence [StacksAlg] (Tag 0BY1) identifies these curves uniquely by the identity of H(x)~\widetilde{\mathcal H(x)}. Call the common curve CxC_x.

Use the inverse limit of connected components of punctured ordinary open neighborhoods to define TxXT_xX. The cofinal branch stars in Corollary 4.28.5 identify this set, for either model, with its residual points. Fix one intrinsic direction, and let p,p′p,p' be its two labels on CxC_x. We first identify these labels without presupposing equality of the two metric parameters.

Every ψ∈k(Cx)×\psi\in k(C_x)^\times lifts to a norm-one analytic unit germ gg at xx: the proof of Proposition 4.24.2 represents it by an affinoid fraction P/QP/Q with P(x)Q(x)≠0P(x)Q(x)\ne 0, and Proposition 2.2.4 makes P,QP,Q invertible near xx. By Lemma 4.33.1, the sign of −log⁡∣g∣-\log|g| on a sufficiently thin collar in the first model is the sign of ord⁡p(ψ)\operatorname{ord}_p(\psi). Cofinality of the two branch-star systems puts every sufficiently thin collar in the second model inside that collar. Its Gauss ray therefore has the same sign, and its slope formula gives sgn⁡ord⁡p(ψ)=sgn⁡ord⁡p′(ψ)(ψ∈k(Cx)×).\operatorname{sgn}\operatorname{ord}_p(\psi) =\operatorname{sgn}\operatorname{ord}_{p'}(\psi) \qquad(\psi\in k(C_x)^\times). Thus the two valuation rings in the common function field are identical. Their centres on the separated proper curve CxC_x are equal, by the closed-diagonal argument in Theorem 4.29.2. Consequently p=p′p=p', and their residue fields, separable degrees and inseparable degrees coincide.

Choose now a uniformizer ψ∈k(Cx)×\psi\in k(C_x)^\times at this common point pp, and lift it to a norm-one unit germ gg. Write e=[κ(p):k]insepe=[\kappa(p):k]_{\mathrm{insep}}. The all-point collar formula gives, on the intersection of sufficiently thin collars, ∣g∣=Rpe=(Rp′)e.|g|=R_p^e=(R'_p)^e. Hence Rp=Rp′R_p=R'_p there. By alternating the two cofinal stars one can choose a collar of each contained in a fixed collar of the other. After increasing both inner radii beyond these bounds, the equality of radius functions gives equality of every thinner collar as a subset of XX. Their analytic structures agree by the analytic-domain universal property. Thus the entire radius germ is intrinsic.

Finally choose a common finite Galois extension rationalizing the residual-point supports in both models. The inverse image of an agreed thin collar is, in either presentation, a finite disjoint union of standard open annuli. These annuli are path connected, so their decomposition is the intrinsic connected component decomposition. Corresponding components are therefore the same analytic domains. On each, the two annular coordinates have the same radius function, namely the pullback of RpR_p. Their identity is a radius-preserving annular isomorphism, and Corollary 4.19.1 identifies the Gauss points of every radius. Projection gives the same Gauss-ray germ on XX, with the same parameter −log⁡r-\log r.

The slope formula is Theorem 4.31.3, now with all its geometric data intrinsic. The finite-support weighted identity is Corollary 4.31.4. All constructions use the given absolute value. ◻

4.34 Equivariant marking over a nondiscrete valuation ring

Throughout this section KK is a complete nontrivially valued field, O=K∘O=K^\circ, and k=K~k=\widetilde K. A split nodal model is a separated flat finitely presented OO-scheme with smooth generic fibre and reduced nodal special fibre, whose nodes are kk-rational and admit the actual rational pointed charts of Theorem 4.18.2. Its formal generic fibre is assumed Hausdorff. A marking is a finite set of distinct KK-rational generic points. In the proper case each extends uniquely to an OO-section by [StacksAlg] (Tag 0BX5).

Lemma 4.34.1. NAG-002144 (An intrinsic blowup for a nodal cluster). Let qq be a node of a split nodal model X\mathcal X, and let a nonempty finite set P(q)P(q) of marked sections specialize to qq. There is a finite-type ideal Iq\mathcal I_q, equal to the unit ideal away from qq and open after formal completion, such that its blowup is again a split nodal model, retains every old component point, and moves at least one member of P(q)P(q) to a smooth special point. No member of P(q)P(q) moves to an outer attaching node. The ideal depends only on the marked cluster and the model, so every semilinear isometric model isomorphism carrying one marked cluster to another carries their ideals to one another. Projectivity over OO is preserved.

Proof. Use a pointed chart st=bst=b at qq, with 0<∣b∣<10<|b|<1. The rational-pointed tube comparison identifies ]q[]q[ with ∣b∣<∣s∣<1|b|<|s|<1. A section specializing to qq lifts uniquely to the pointed auxiliary chart: its pullback along that chart is étale over the henselian ring OO and has the specified rational closed point, so the henselian property of Lemma 4.15.1 supplies the lift. Write aP=s(P),dP=t(P),aPdP=b,δq=max⁡P∈P(q)max⁡(∣aP∣,∣dP∣).a_P=s(P),\qquad d_P=t(P),\qquad a_Pd_P=b,\qquad \delta_q=\max_{P\in P(q)}\max(|a_P|,|d_P|). Then 0<δq<10<\delta_q<1, δq∈∣K×∣\delta_q\in|K^\times|, and ∣b∣≤δq2|b|\leq\delta_q^2. Choose λ∈O\lambda\in O with ∣λ∣=δq|\lambda|=\delta_q. Near qq put Iq=(λ)+Fitt⁡1ΩX/O.\mathcal I_q=(\lambda)+ \operatorname{Fitt}_1\Omega_{\mathcal X/O}. Shrink this neighborhood to omit every other node. The construction and proof of Lemma 4.27.2 apply verbatim to these rational pointed charts. That proof uses algebraic closedness only to supply the rational split chart, which is a hypothesis here: finite generation, gluing with the unit ideal, étale descent of the charts, retention of old vertices, and projectivity are all proved over the displayed valuation ring. Its local pullback is (λ,s,t)(\lambda,s,t), so Lemma 4.27.1 gives the actual three charts.

Every section in the cluster lies in the λ\lambda-chart: its coordinates there are u=aP/λu=a_P/\lambda, v=dP/λv=d_P/\lambda, both in OO. At least one section has uu or vv a unit by the definition of δq\delta_q. If ∣b∣=δq2|b|=\delta_q^2, the entire middle special chart uv=b/λ2‾uv=\overline{b/\lambda^2} is smooth. If ∣b∣<δq2|b|<\delta_q^2, its only node is u=v=0u=v=0. Thus at least one marked section becomes smooth, and the only possible remaining nodal marked specialization is the middle node. The outer attaching nodes are outside the middle chart.

It remains to check that the centre is intrinsic. Between two normalized annular coordinates A(∣b∣,1)A(|b|,1) and A(∣b′∣,1)A(|b'|,1), Corollary 4.19.1 gives ∣b′∣=∣b∣|b'|=|b| and either ∣s′∣=∣s∣|s'|=|s| or ∣s′∣=∣b∣/∣s∣|s'|=|b|/|s|. It therefore preserves the unordered pair (∣aP∣,∣dP∣)(|a_P|,|d_P|) and hence δq\delta_q. This also holds semilinearly, since the field automorphism is isometric. A principal ideal of OO depends only on the absolute value of its nonzero generator, and the Fitting ideal is intrinsic. The glued ideal is consequently independent of all choices and has the asserted covariance. ◻

Lemma 4.34.2. NAG-002145 (An intrinsic blowup for a smooth cluster). Let at least two distinct marked sections of a split nodal model X\mathcal X specialize to the same smooth point qq. There is an intrinsic finite-type ideal supported at qq whose blowup leaves all these sections smooth and splits their specializations into at least two nonempty classes. It has the same semilinear covariance, retention of old vertices, and projectivity properties as Lemma 4.34.1.

Proof. The point qq is rational, being the special point of a section. Choose an algebraic smooth coordinate TT at qq, and write aP=T(P)a_P=T(P) for the finitely many sections in its cluster. All aPa_P have the same residue. The tube coordinate is injective on its KK-points by Corollary 4.17.2, so these values are pairwise distinct. Put 0<δq=max⁡P,Q∣aP−aQ∣<1,∣λ∣=δq.0<\delta_q=\max_{P,Q}|a_P-a_Q|<1,\qquad |\lambda|=\delta_q . Choose one section S0S_0 in the cluster, and define Iq=(λ)+IS0.\mathcal I_q=(\lambda)+\mathcal I_{S_0}. The section is contained in the smooth locus: both its generic and special points are smooth, and the spectrum of the rank-one valuation ring has just those two points. Its image is an effective Cartier divisor. Indeed it is finite flat of rank one over OO, and [StacksAlg] (Tag 0B9D) applies in the relative smooth curve locus; away from the image its ideal is the unit ideal. Thus Iq\mathcal I_q is a finite-type quasi-coherent ideal. It is the unit ideal away from qq, since λ\lambda is invertible on the generic fibre and S0S_0 has only the one special point. Its completion is open.

Near qq, the ideal of every section SPS_P in the cluster is generated by T−aPT-a_P. To see this, choose a Cartier generator fPf_P of its ideal. The function T−aPT-a_P belongs to that ideal, so equals fPhPf_Ph_P. On the smooth special fibre both f‾P\overline f_P and T−aP‾\overline{T-a_P} are local parameters at qq; the former assertion follows also from the relative Cartier statement in Tag 0B9D. Hence hP(q)≠0h_P(q)\ne 0, and hPh_P is a unit after shrinking. Since aP−aQ∈λOa_P-a_Q\in\lambda O for every pair, (λ,T−aP)=(λ,T−aQ).(\lambda,T-a_P)=(\lambda,T-a_Q). These ideals are the unit ideal away from qq. Consequently the global centre is independent of S0S_0. Changing smooth coordinates does not change δq\delta_q: an open-unit-disc isomorphism has linear coefficient of norm one and all higher coefficients of norm at most one, so for ∣a∣,∣a′∣<1|a|,|a'|<1 its difference divided by a−a′a-a' is a unit of norm one. This is the coefficient calculation in Theorem 4.21.2. Thus the centre is also independent of the coordinate, and is covariant under the indicated semilinear isomorphisms.

Apply Lemma 4.27.3 with centre aP0a_{P_0} and scale λ\lambda, reusing its actual blowup charts. Every marked section belongs to the smooth λ\lambda-chart, with coordinate U=(T−aP0)/λU=(T-a_{P_0})/\lambda. At least two of these coordinates have distinct residues, because one pairwise difference has norm exactly δq\delta_q. All marked specializations therefore remain smooth, and the old cluster separates into at least two nonempty residue classes. ◻

Theorem 4.34.3. NAG-002146 (Equivariant finite marked refinement). Let X/O\mathcal X/O be a proper split nodal model, and let PP be a finite set of distinct KK-rational points of its generic fibre. A finite sequence of algebraic blowups with open completed centres produces a split nodal model X′\mathcal X' in which the marked sections have pairwise distinct smooth specializations. Every old component point is retained. If X\mathcal X is projective, so is X′\mathcal X'. If a finite group GG acts semilinearly and isometrically on X\mathcal X and preserves PP, the sequence can be chosen GG-equivariant, with the given generic action unchanged.

Proof. First treat all nodes containing marked specializations. For each such node take the intrinsic ideal of Lemma 4.34.1, and blow up the product of these finitely many ideals. In a neighborhood of one centre that omits the other centres, every other ideal is the unit ideal. Thus this simultaneous blowup has exactly the local charts already proved there. This argument uses a finite product of finite-type ideals, not an intersection. The number of marked sections meeting nodes strictly decreases: at each treated node at least one becomes smooth, while every previously smooth section is untouched. After at most ∣P∣|P| rounds all marked sections are smooth.

Now treat every smooth point containing at least two marked specializations, again by blowing up the finite product of the ideals in Lemma 4.34.2. All marks stay smooth. The partition of PP by specializations strictly refines whenever a collision is present, and different old classes never merge because their centres are distinct and each blowup is an isomorphism off its centre. A partition of a finite set has at most ∣P∣|P| classes, so this second process also terminates after finitely many rounds. Each blowup preserves properness, split nodality, finite presentation, old component points, and projectivity when present, by the two preceding lemmas.

The group permutes the relevant marked clusters at every round and preserves their intrinsic ideals. It therefore preserves their product. An automorphism of a scheme preserving an ideal acts on its graded Rees algebra and hence on its relative Proj. The multiplication and cocycle identities are inherited from the original action on that algebra; semilinear base automorphisms are handled by the same graded construction. Thus every stage is GG-equivariant. No integer-valued contact function or discreteness of the value group enters either termination argument. ◻

Corollary 4.34.4. NAG-002147 (Marked curves of genus at least two). Let KK be complete of characteristic zero with its given nontrivial non-Archimedean absolute value, let C/KC/K be a smooth projective geometrically connected curve of genus at least two, and let DD be a finite reduced closed subscheme of CC. There is a finite Galois extension L/KL/K and a projective split nodal model of CLC_L whose LL-rational marked sections DLD_L have distinct smooth specializations, with the canonical Galois action extending to the model. Consequently (C∖D)an(C\setminus D)^{\mathrm{an}} admits a jointly continuous strong deformation retraction onto a finite metric graph with one infinite ray for every closed point of DD, using the original absolute value.

Proof. Corollary 4.14.3 supplies a projective stable model after a finite extension. Enlarge to a finite Galois extension splitting the finitely many residue fields of DD. Characteristic zero makes those extensions separable, so the normal closure remains finite Galois. Apply Theorem 4.18.2 to the finitely many special nodes, and then take a finite Galois closure over KK once more. Base change of the stable model remains stable and projective. Rational split pointed charts persist under this further extension. The new nodal fibre is still geometrically reduced.

The canonical Galois action extends to this stable model by Corollary 4.14.5. Apply Theorem 4.34.3 to the finite invariant set of now rational markings. The resulting marked model remains projective and Galois-compatible. The compact skeleton exists over the final field even when its residue field is imperfect, by Corollary 4.28.5. Theorem 4.32.2 supplies the entire equivariant punctured homotopy. Corollary 4.9.3 descends it to the original analytic curve; the open-quotient product property of Theorem 4.9.2 applies to this noncompact space. The quotient-ray and length calculation is precisely the one in Corollary 4.32.3, whose descent argument uses only the equivariant marked model, not the discreteness used there to construct it. ◻

4.35 Equivariant genus-zero models

Lemma 4.35.1. NAG-002148 (Gauss discs and fractional linear maps). Let FF be a complete nontrivially valued field. For a∈Fa\in F and r>0r>0, let ζa,r\zeta_{a,r} be the Gauss seminorm with centre aa and radius rr. The set BF={ζa,r:a∈F, r>0}\mathcal B_F=\{\zeta_{a,r}:a\in F,\ r>0\} has the following tree metric: if x=ζa,rx=\zeta_{a,r}, y=ζb,sy=\zeta_{b,s}, and R=max⁡(r,s,∣a−b∣)R=\max(r,s,|a-b|), then d(x,y)=2log⁡R−log⁡r−log⁡s.d(x,y)=2\log R-\log r-\log s. Their interval is obtained by increasing the first radius from rr to RR and the second from ss to RR, with the common endpoint ζa,R=ζb,R\zeta_{a,R}=\zeta_{b,R} identified. Every finite hull is a finite compact metric tree, embedded with its metric topology in (PF1)an(\mathbf P^1_F)^{\mathrm{an}}. Fractional linear maps and semilinear maps induced by isometric field automorphisms preserve BF\mathcal B_F, its metric, and its intervals. They carry points with radius in ∣F×∣|F^\times| to points of that same kind.

Proof. The weighted Gauss formula gives ζa,r=ζb,s ⟺ r=s and ∣a−b∣≤r.\zeta_{a,r}=\zeta_{b,s} \ \Longleftrightarrow\ r=s\ \hbox{and}\ |a-b|\leq r. For the forward implication evaluate T−aT-a and T−bT-b. For the converse expand a polynomial about either centre: translation by an element of norm at most rr is contractive for the radius-rr Gauss norm, as is its inverse. Define ζa,r≤ζb,s\zeta_{a,r}\leq\zeta_{b,s} by r≤sr\leq s and ∣a−b∣≤s|a-b|\leq s. The displayed equality shows that this is well defined and that the least common upper bound is the point of radius RR above.

For finitely many such points choose one upper bound ζ0,R0\zeta_{0,R_0}. Their upward intervals to that point intersect in terminal intervals, beginning at their pairwise least upper bounds. They therefore form a finite tree: after subdividing at those finitely many upper bounds, each added interval meets the preceding tree in a terminal interval and cannot create a circuit. Remove any terminal part beyond the last join to obtain the hull. Assign length log⁡(v/u)\log(v/u) to a radial interval u≤r≤vu\leq r\leq v. The unique path just described has the stated length, proving the metric formula and tree assertion. Each radial map is continuous by the Gauss formula, and is injective because evaluation of its centred coordinate recovers rr. The finite tree maps continuously and injectively into the Hausdorff analytic projective line. Compactness then identifies its metric and subspace topologies.

Translations and nonzero scalings plainly preserve the formula. Inversion I(T)=1/TI(T)=1/T satisfies I(ζa,r)={ζa−1, r/∣a∣2,r<∣a∣,ζ0, r−1,r≥∣a∣.I(\zeta_{a,r})= \begin{cases} \zeta_{a^{-1},\,r/|a|^2},&r<|a|,\\ \zeta_{0,\,r^{-1}},&r\geq|a|. \end{cases} In the second case ζa,r=ζ0,r\zeta_{a,r}=\zeta_{0,r}, and reversing the coefficients of TnP(1/T)T^nP(1/T) proves the formula. For r<∣a∣r<|a|, write T=a+UT=a+U and expand (a+U)−1−a−1(a+U)^{-1}-a^{-1}. Its linear term has norm r/∣a∣2r/|a|^2, and each higher term has strictly smaller norm on the radius-rr Gauss algebra. This substitution is contractive into the algebra of radius r/∣a∣2r/|a|^2 in the inverse coordinate; the inverse substitution has the same property. Applying both inequalities to each polynomial proves equality of the Gauss norms and the formula.

On a radial interval the first formula preserves logarithmic lengths until r=∣a∣r=|a|; the second reverses the logarithmic parameter afterwards. If both parts occur, their images meet only at ζ0,∣a∣−1\zeta_{0,|a|^{-1}}, since their smaller discs have centres separated by ∣a∣−1|a|^{-1}. Thus no backtracking occurs, and inversion maps every interval isometrically to the interval between its endpoints. Every fractional linear map is a composition of affine maps and inversion, so it preserves the metric and the intervals. An isometric field automorphism sends ζa,r\zeta_{a,r} to ζσ(a),r\zeta_{\sigma(a),r}, which has the same properties. The displayed formulas also preserve radii in the value group. ◻

Lemma 4.35.2. NAG-002149 (A fixed Gauss point after a finite extension). Suppose a finite group acts on (PF1)an(\mathbf P^1_F)^{\mathrm{an}} by semilinear fractional linear maps with isometric coefficient automorphisms. The hull of the orbit of ζ0,1\zeta_{0,1} has a unique metric centre cc and the group fixes cc. One may write c=ζa,rc=\zeta_{a,r} with a∈Fa\in F and r2∈∣F×∣r^2\in|F^\times|. After adjoining a square root of an element of norm r2r^2, the point is a rational-radius Gauss point over that finite extension.

Proof. The preceding lemma gives a finite invariant compact metric tree TT. Its leaves belong to the original finite orbit, since the hull has no extra terminal interval. Choose diameter endpoints A,BA,B, let the diameter be dd, and let cc be the midpoint of [A,B][A,B]. For any z∈Tz\in T, let yy be its attaching point on [A,B][A,B], put t=d(A,y)t=d(A,y) and h=d(y,z)h=d(y,z). The diameter inequalities give h≤min⁡(t,d−t)h\leq\min(t,d-t). Hence d(c,z)=∣t−d/2∣+h≤d/2.d(c,z)=|t-d/2|+h\leq d/2. For x≠cx\ne c, the tree paths give max⁡(d(x,A),d(x,B))=d/2+d(x,c)>d/2\max(d(x,A),d(x,B))=d/2+d(x,c)>d/2. Thus cc is the unique point minimizing the maximum distance to TT, so every isometry of TT fixes it.

Write A=ζα,uA=\zeta_{\alpha,u} and B=ζβ,vB=\zeta_{\beta,v}, where u,v∈∣F×∣u,v\in|F^\times|, and put R=max⁡(u,v,∣α−β∣)R=\max(u,v,|\alpha-\beta|). All these positive numbers belong to ∣F×∣|F^\times|. If cc is on the radial part from AA to the join, then c=ζα,rc=\zeta_{\alpha,r} with r2=R2u/vr^2=R^2u/v; on the other part it has centre β\beta and r2=R2v/ur^2=R^2v/u. If the orbit is a singleton, simply take c=ζ0,1c=\zeta_{0,1}. Choose γ∈F×\gamma\in F^\times with ∣γ∣=r2|\gamma|=r^2. In the finite extension obtained by adjoining γ\sqrt\gamma, the uniquely extended absolute value gives ∣γ∣=r|\sqrt\gamma|=r. All formulas for the action and for Gauss points commute with scalar extension, so the lifted point remains fixed under the extended action whenever that action is specified. ◻

Lemma 4.35.3. NAG-002150 (The integral stabilizer of the unit Gauss point). An element of PGL⁡2(F)\operatorname{PGL}_2(F) fixes ζ0,1\zeta_{0,1} if and only if it is represented by a matrix in GL⁡2(F∘)\operatorname{GL}_2(F^\circ). The same criterion holds for a semilinear fractional linear transformation with isometric coefficient automorphism. Such a transformation extends uniquely to a semilinear automorphism of PF∘1\mathbf P^1_{F^\circ}.

Proof. A matrix M=(abcd)M=\left(\begin{smallmatrix}a&b\\c&d\end{smallmatrix}\right) in GL⁡2(F∘)\operatorname{GL}_2(F^\circ) reduces to an invertible matrix over the residue field. Its denominator cT+dcT+d has Gauss norm one. If a polynomial PP has Gauss norm one, its homogenized substitution has integral coefficients and nonzero reduction, because the reduced fractional transformation is an automorphism of the rational function field. Thus P((aT+b)/(cT+d))P((aT+b)/(cT+d)) again has Gauss norm one. Scaling an arbitrary nonzero polynomial by one of its maximal coefficients proves that MM fixes the Gauss point.

Conversely rescale a representing matrix so that the largest entry norm is one, and write Δ=ad−bc\Delta=ad-bc. Suppose first max⁡(∣c∣,∣d∣)=1\max(|c|,|d|)=1. If necessary precompose by inversion, which fixes ζ0,1\zeta_{0,1}, so that ∣d∣=1|d|=1. Then aT+bcT+d=bd+Δd2T1+(c/d)T.\frac{aT+b}{cT+d} =\frac bd+\frac{\Delta}{d^2} \frac{T}{1+(c/d)T}. The last fractional transformation is represented by an integral matrix of determinant one and fixes the Gauss point by the first part. Hence the image of ζ0,1\zeta_{0,1} is ζb/d,∣Δ∣\zeta_{b/d,|\Delta|}. It equals ζ0,1\zeta_{0,1} only when ∣Δ∣=1|\Delta|=1. If max⁡(∣c∣,∣d∣)<1\max(|c|,|d|)<1, the numerator row has maximum norm one and ∣Δ∣<1|\Delta|<1. Postcomposing by inversion interchanges the two rows, reducing to the previous case and showing that the image cannot be the Gauss point. Thus a fixing matrix has ∣Δ∣=1|\Delta|=1, as required.

An isometric coefficient automorphism fixes ζ0,1\zeta_{0,1}, so it does not change the criterion. An integral invertible matrix defines an automorphism of PF∘1\mathbf P^1_{F^\circ} by its homogeneous linear coordinates. Two normalized representatives of the same projective matrix differ by a unit and define the same map. Uniqueness of an extension follows either on the two standard affine charts or by separation and density of the generic fibre. In particular, generic group identities hold for the extended maps. ◻

Theorem 4.35.4. NAG-002151 (A Galois-compatible genus-zero model). Let KK be a complete nontrivially valued field of characteristic zero, let C/KC/K be a smooth projective geometrically connected genus-zero curve, and let D⊂CD\subset C be finite and reduced. There is a finite Galois extension M/KM/K, with its uniquely extended absolute value, and an identification CM≃PM1C_M\simeq\mathbf P^1_M for which the canonical semilinear Galois action extends to PM∘1\mathbf P^1_{M^\circ}. After equivariant admissible algebraic blowups, this model can be chosen split nodal and projective, with all points of DMD_M rational and specializing to distinct smooth points.

Proof. Proposition 4.30.1 gives a finite splitting field for CC. Enlarge it to a finite Galois extension L/KL/K also splitting the residue fields of DD, and choose CL≃PL1C_L\simeq\mathbf P^1_L. In this coordinate the canonical Galois action is by semilinear fractional linear transformations. For completeness, any linear automorphism of the projective line has this form: the pullback of the affine coordinate has exactly one simple pole, at the rational inverse image of infinity; a rational function with that pole divisor is a ratio of linear polynomials. Composing a semilinear map with the inverse coefficient automorphism reduces it to this case.

Apply Lemma 4.35.2 to this finite action. It gives a fixed point c=ζa,rc=\zeta_{a,r}, with a∈La\in L and r2∈∣L×∣r^2\in|L^\times|. Choose γ∈L×\gamma\in L^\times with ∣γ∣=r2|\gamma|=r^2, and let M/KM/K be a finite Galois extension containing L(γ)L(\sqrt\gamma). Such a finite Galois closure exists because the characteristic is zero. Set λ=γ\lambda=\sqrt\gamma in MM. The canonical lift cM=ζa,rc_M=\zeta_{a,r} is fixed by Gal⁡(M/K)\operatorname{Gal}(M/K): an automorphism restricts to one of Gal⁡(L/K)\operatorname{Gal}(L/K), and its matrix has coefficients in LL; the explicit Gauss-disc formulas of Lemma 4.35.1 commute with this extension. Automorphisms trivial on LL fix both the centre and the radius.

Replace the coordinate by U=(T−a)/λU=(T-a)/\lambda. The fixed point becomes the unit Gauss point. Lemma 4.35.3 extends every semilinear Galois transformation to PM∘1\mathbf P^1_{M^\circ}, and its uniqueness gives the group identities. This is a projective smooth model with geometrically integral special fibre. The finite rational marking set remains invariant. Apply Theorem 4.34.3 to obtain the asserted projective split marked model. ◻

Corollary 4.35.5. NAG-002152 (Genus-zero retraction over the original field).

For K,C,DK,C,D as in Theorem 4.35.4, (C∖D)an(C\setminus D)^{\mathrm{an}} admits a jointly continuous strong deformation retraction onto a finite metric graph with one infinite ray for each closed point of DD. If DD is empty, the image can be a single point. All metrics use the original absolute value on KK.

Proof. Apply Theorem 4.32.2 to the Galois-compatible marked model just constructed. Its smooth residual tubes are covered by Corollary 4.28.5, including imperfect residue fields. The entire homotopy is equivariant, so Corollary 4.9.3 descends it. The finite metric quotient and the ray-orbit calculation are those already established in Corollary 4.32.3. When DD is empty no marked refinement is needed: the good model PM∘1\mathbf P^1_{M^\circ} has its single fixed Gauss vertex as skeleton. ◻

4.36 An explicit genus-one semistable model

Lemma 4.36.1. NAG-002167 (Short Weierstrass equations and their parameter). Let FF be a field of characteristic zero and let (E,OE)(E,O_E) be a smooth projective geometrically connected genus-one curve with an FF-rational point. It has a pointed plane equation y2=x3+Ax+B,OE=(0:1:0),4A3+27B2≠0.y^2=x^3+Ax+B,\qquad O_E=(0:1:0),\qquad 4A^3+27B^2\ne0 . The value j=1728 4A34A3+27B2j=1728\,\frac{4A^3}{4A^3+27B^2} is independent of the pointed short equation. Two such equations with the same jj become pointed isomorphic after a finite extension of FF.

Proof. Geometric connectedness and smoothness imply geometric integrality, and [StacksAlg] (Tag 04L2(4)) gives H0(E,OE)=FH^0(E,\mathcal O_E)=F. The curve is regular and hence Gorenstein by [StacksAlg] (Tags 056S and 0DWG). The Riemann–Roch and duality formulas [StacksAlg] (Tag 0BS6) give deg⁡ωE=0\deg\omega_E=0 and h0(E,OE(nOE))=n(n≥1).h^0(E,\mathcal O_E(nO_E))=n \qquad(n\geq1). Indeed the dual term has degree −n-n and no nonzero section. A nonzero section of an invertible sheaf injects OE\mathcal O_E with a finite-length quotient, so its degree is the nonnegative length of that quotient; this excludes negative degree. Choose functions x,yx,y with exact pole orders two and three at OEO_E and no other poles. The six functions 1,x,y,x2,xy,x31,x,y,x^2,xy,x^3 have distinct pole orders and form a basis of H0(E,OE(6OE))H^0(E,\mathcal O_E(6O_E)). Consequently y2y^2 is their linear combination, with a nonzero x3x^3 coefficient. Scaling both x,yx,y by that coefficient makes it one. This gives y2+a1xy+a3y=x3+a2x2+a4x+a6.y^2+a_1xy+a_3y =x^3+a_2x^2+a_4x+a_6 . The linear system ∣3OE∣|3O_E|, with basis 1,x,y1,x,y, is very ample by [StacksAlg] (Tag 0H2V). Its image is therefore the indicated plane cubic. Indeed the image is an integral plane curve, and its homogeneous ideal is principal by [StacksAlg] (Tag 0BYB). If its homogeneous generator has degree ee, the genus formula [StacksAlg] (Tag 0BYD) gives (e−1)(e−2)/2=1(e-1)(e-2)/2=1, whence e=3e=3. The cubic already displayed lies in this ideal and is nonzero, so it is a scalar multiple of the generator. Completing the square in yy and eliminating the x2x^2 term by translation gives the stated short equation. These coordinate changes fix (0:1:0)(0:1:0). The partial derivatives of the short equation show smoothness precisely when 4A3+27B2≠04A^3+27B^2\ne 0.

A pointed isomorphism between short equations pulls back the second xx to αx+r\alpha x+r and the second yy to βy+sx+t\beta y+sx+t, since it preserves the pole-order filtration, with αβ≠0\alpha\beta\ne 0. Comparing the coefficients of xy,y,x3,x2xy,y,x^3,x^2 gives s=t=r=0s=t=r=0 and β2=α3\beta^2=\alpha^3. Thus with u=β/αu=\beta/\alpha one has α=u2\alpha=u^2, β=u3\beta=u^3, A′=u4AA'=u^4A, and B′=u6BB'=u^6B. This preserves jj. Conversely, for j≠0,1728j\ne 0,1728 put r=A′/Ar=A'/A, s=B′/Bs=B'/B. Equality of jj gives r3=s2r^3=s^2. After choosing uu with u2=s/ru^2=s/r, one has u4=ru^4=r and u6=su^6=s, giving the isomorphism. For j=0j=0 take a sixth root of B′/BB'/B; for j=1728j=1728 take a fourth root of A′/AA'/A. All these extensions are finite. ◻

Lemma 4.36.2. NAG-002168 (The integral plane-cubic family). Let O=F∘O=F^\circ for a complete nontrivially valued field. For coefficients ai∈Oa_i\in O, the projective cubic Y2Z+a1XYZ+a3YZ2=X3+a2X2Z+a4XZ2+a6Z3Y^2Z+a_1XYZ+a_3YZ^2 =X^3+a_2X^2Z+a_4XZ^2+a_6Z^3 is projective, flat and finitely presented over OO, and its section (0:1:0)(0:1:0) is relatively smooth. If its generic fibre is smooth and its geometric special fibre is smooth or nodal, it is a nodal model of that generic curve.

Proof. The homogeneous coordinate ring is a free O[Y,Z]O[Y,Z]-module with basis 1,X,X21,X,X^2, because the equation is monic of degree three in XX. It is therefore OO-flat. Localization at each homogeneous coordinate remains flat; its degree-zero submodule is a direct summand and hence flat. These degree-zero rings are the standard affine charts of the projective cubic. The single homogeneous equation gives finite presentation, and the closed embedding in PO2\mathbf P^2_O gives projectivity. In the affine chart Y=1Y=1, the partial derivative with respect to ZZ is one at (X,Z)=(0,0)(X,Z)=(0,0), so the indicated section lies in the relative smooth locus. A nontrivial real-valued valuation ring has only generic and closed points. The stated two fibre conditions thus verify the nodal-model hypotheses. ◻

Proposition 4.36.3. NAG-002169 (The Legendre model when two is a unit). Let FF be complete of characteristic zero with nontrivial absolute value and residue characteristic different from two. Every pointed smooth projective genus-one curve over FF has, after a finite extension, a projective nodal model over the resulting valuation ring. Its special fibre is either smooth or an irreducible cubic with one ordinary double point.

Proof. Use Lemma 4.36.1 and split its cubic as (x−e1)(x−e2)(x−e3)(x-e_1)(x-e_2)(x-e_3) after a finite extension. The roots are distinct. Choose the pair labelled e1,e2e_1,e_2 to have maximal pairwise distance. After scaling x−e1x-e_1 by e2−e1e_2-e_1 and adjoining a square root for the corresponding yy-scaling, the equation becomes Y2=X(X−1)(X−λ),λ=e3−e1e2−e1,∣λ∣≤1,∣1−λ∣≤1.Y^2=X(X-1)(X-\lambda),\qquad \lambda=\frac{e_3-e_1}{e_2-e_1},\qquad |\lambda|\leq1,\quad|1-\lambda|\leq1 . At least one of the last two norms is one. If both are one, the three residual roots are distinct. Since two is a unit, the affine partial derivatives then have no common zero on the cubic. The point at infinity is smooth as in Lemma 4.36.2.

If ∣λ∣<1|\lambda|<1, the residual equation is v2=u2(u−1)v^2=u^2(u-1), with its sole singular point at the origin. There the derivative ideal is (u(2−3u),2v)=(u,v)(u(2-3u),2v)=(u,v). The local hypersurface ring has depth one: uu is a nonzerodivisor since the equation is not divisible by uu, and its dimension is one. Its first differential Fitting ideal is the derivative ideal just computed. Its residue field is the base field, so [StacksAlg] (Tag 0C4D(5)) proves that this point is a node. If ∣1−λ∣<1|1-\lambda|<1, translate u=X−1u=X-1; the same computation uses v2=u2(u+1)v^2=u^2(u+1) and derivative u(−2−3u)u(-2-3u). The quadratic terms have two distinct linear factors over a separable extension, because the characteristic is not two. The residual cubic is irreducible over the algebraic closure: in the first case u−1u-1 is not a square in k‾(u)\overline k(u), so v2−u2(u−1)v^2-u^2(u-1) is irreducible, and the second case is identical with u+1u+1. The integral-family lemma now supplies the required model. ◻

Proposition 4.36.4. NAG-002170 (The Deuring model in residue characteristic two). Let FF be complete of characteristic zero with nontrivial absolute value and residue characteristic two. Every pointed smooth projective genus-one curve over FF has, after a finite extension, a projective model of the form y2+axy+y=x3,a∈O.y^2+a xy+y=x^3,\qquad a\in O. Its special fibre is smooth if ∣j∣≤1|j|\leq 1 and has exactly one ordinary double point if ∣j∣>1|j|>1. Here jj is the parameter of Lemma 4.36.1, and OO is the valuation ring after extension.

Proof. Put b=a3b=a^3. Completing the square and eliminating the quadratic term over the characteristic-zero field gives the short coefficients A=−a(b−24)48,B=b2−36b+216864.A=-\frac{a(b-24)}{48},\qquad B=\frac{b^2-36b+216}{864}. Expansion gives the polynomial identity b(b−24)3−(b2−36b+216)2=1728(b−27).b(b-24)^3-(b^2-36b+216)^2=1728(b-27). Hence the discriminant and parameter of this equation are Δ=−16(4A3+27B2)=b−27,j=b(b−24)3b−27.\Delta=-16(4A^3+27B^2)=b-27,\qquad j=\frac{b(b-24)^3}{b-27}.

If ∣j∣≤1|j|\leq 1, choose a root of the monic polynomial Pj(B)=B(B−24)3−j(B−27).P_j(B)=B(B-24)^3-j(B-27). It belongs to a finite extension. Every such root bb has norm at most one: at a value greater than one the monic highest term would strictly dominate all other terms. Modulo the maximal ideal, residue characteristic two gives P‾j(1)=1\overline P_j(1)=1. Thus b‾≠1\overline b\ne 1 and ∣b−27∣=1|b-27|=1. Adjoin a cube root aa of bb; then aa is integral. The discriminant is nonzero and the displayed parameter equals the prescribed jj. The short-equation lemma therefore supplies an isomorphism to the original pointed curve after one more finite extension.

If ∣j∣>1|j|>1, put ϵ=j−1\epsilon=j^{-1} and solve b=27+ϵb(b−24)3.b=27+\epsilon b(b-24)^3. The right side is a contraction of the closed unit ball: the polynomial B(B−24)3B(B-24)^3 has integral coefficients and is one-Lipschitz on that ball, while ∣ϵ∣<1|\epsilon|<1. Iteration starting at 2727 converges in FF to a solution. It has b‾=1\overline b=1, so both bb and b−24b-24 are units. The equation then gives 0<∣b−27∣=∣ϵ∣<1.0<|b-27|=|\epsilon|<1 . Adjoin a cube root aa and then the finite short-equation isomorphism as above. In either case further extension preserves all the strict and unit norm conditions.

We verify the special fibre directly. At infinity it is smooth by Lemma 4.36.2. On the affine chart in characteristic two, let α=a‾\alpha=\overline a. The partial derivatives are Fx=αy+x2,Fy=αx+1.F_x=\alpha y+x^2,\qquad F_y=\alpha x+1. If α=0\alpha=0, there is no singular point. If α≠0\alpha\ne 0, the only candidate is x0=α−1x_0=\alpha^{-1}, y0=α−3y_0=\alpha^{-3}. Substitution into the equation shows that it is singular exactly when α3=1\alpha^3=1. Thus the integral-jj case is geometrically smooth. In the other case use u=x−x0u=x-x_0, v=y−y0v=y-y_0. The derivative ideal is (αv+u2,αu)=(u,v).(\alpha v+u^2,\alpha u)=(u,v). The local hypersurface has depth one since uu is a nonzerodivisor: its equation modulo uu is v2v^2, so the equation is not divisible by uu. Its residue point is rational and its first differential Fitting ideal is the displayed maximal ideal. By [StacksAlg] (Tag 0C4D(5)) it is a node. Equivalently, its quadratic part v2+αuv+α−1u2,v^2+\alpha uv+\alpha^{-1}u^2, which has two distinct factors after a finite separable residue extension because α≠0\alpha\ne 0. These calculations exhaust all geometric singular points. Thus the special fibre is geometrically reduced and at worst nodal. The integral-family lemma finishes the model construction. ◻

Theorem 4.36.5. NAG-002171 (Marked genus-one models after finite extension). Let KK be complete of characteristic zero with its given nontrivial non-Archimedean absolute value. Let C/KC/K be a smooth projective geometrically connected genus-one curve, and let D⊂CD\subset C be finite and reduced. There is a finite Galois extension L/KL/K and a projective split nodal model of CLC_L such that all points of DLD_L are rational and have distinct smooth specializations. Consequently (CL∖DL)an(C_L\setminus D_L)^{\mathrm{an}} has a jointly continuous strong deformation retraction onto a finite metric graph with one infinite ray for every point of DLD_L.

Proof. A nonempty affine open of CC has a closed point of finite residue degree by [StacksAlg] (Tag 00FV). After that finite extension the curve has a rational point. Flat cohomology base change [StacksAlg] (Tag 02KH) preserves genus one; geometric integrality is preserved as well. Apply Proposition 4.36.3 or Proposition 4.36.4 according to the residue characteristic.

Take a finite Galois closure over KK that also splits the finitely many residue fields of DD. The constructed projective nodal model survives this base change: flatness and finite presentation are preserved, and the geometric special fibres remain nodal. Apply Theorem 4.18.2 to its finite node set, then take a further finite Galois closure over KK. The actual rational split charts persist. Theorem 4.34.3, used without a group-action hypothesis, separates the finite rational marking set by admissible algebraic blowups. Projectivity is preserved throughout. Finally Theorem 4.32.2, using its characteristic-zero arbitrary-residue branch, gives the asserted homotopy over the final field. Every extension carries the unique absolute value extending the originally specified one. ◻

4.37 Equivariant genus-one models by doubling

We use stable reduction of an auxiliary nodal curve and then separate its two labelled generic components. The local normalization calculation works over a valuation ring without a discreteness assumption.

Lemma 4.37.1. NAG-002204 (The stable double of a genus-one curve). Let KK be a characteristic-zero field and let C/KC/K be a smooth projective geometrically connected curve of genus one. There is a stable curve Y/KY/K of genus d+1≥2d+1\geq 2 whose normalization is the disjoint union of two labelled copies C1,C2C_1,C_2 of CC. The construction is defined over KK; consequently every scalar-extension Galois action preserves the two labels.

Proof. Choose a closed point q∈Cq\in C with residue field EE finite over KK, using [StacksAlg] (Tag 00FV) on a nonempty affine open. Put d=[E:K]d=[E:K]. The extension is separable, since KK has characteristic zero. Glue the two copies of qq by the identity on Spec⁡E\operatorname{Spec}E. This is the pushout of Spec⁡E⨿Spec⁡E⟶C1⨿C2,Spec⁡E⨿Spec⁡E⟶Spec⁡E.\operatorname{Spec}E\amalg\operatorname{Spec}E \longrightarrow C_1\amalg C_2, \qquad \operatorname{Spec}E\amalg\operatorname{Spec}E \longrightarrow\operatorname{Spec}E. The first map is a closed immersion and the second is finite. An affine neighborhood of qq in each copy supplies the affine-neighborhood condition of [StacksAlg] (Tag 0E25), so the pushout is a scheme. It is separated and of finite type over KK by [StacksAlg] (Tags 0E26, 0E27). The natural map ν:C1⨿C2→Y\nu:C_1\amalg C_2\to Y is finite and surjective: on the glued affine chart it is the inclusion R=A×EA ⟶ A⊕A,A⊕A=R(1,0)+R(0,1).R=A\times_E A\ \longrightarrow\ A\oplus A, \qquad A\oplus A=R(1,0)+R(0,1). Properness of C1⨿C2C_1\amalg C_2 and [StacksAlg] (Tag 03GN) show that YY is proper. The fibre-product description commutes with field extensions, since tensoring over KK is exact. Over an algebraic closure the construction therefore glues the two smooth connected copies at dd distinct pairs of rational points. At each glued point the completed local ring is K‾[[u]]×K‾K‾[[v]]≃K‾[[u,v]]/(uv),\overline K[[u]]\times_{\overline K}\overline K[[v]] \simeq\overline K[[u,v]]/(uv), by [StacksAlg] (Tag 0E36) and the smooth rational-point completion calculation already used in Lemma 4.21.1; [StacksAlg] (Tag 0CBW) identifies these as split nodes. Thus YY is geometrically connected and nodal. The finite map ν\nu is an isomorphism off the glued points and has normal source, so it is the normalization: an element of the common total function ring integral over a target affine ring is also integral over its finite normal source ring, and hence belongs to the latter.

The ring description gives the exact sequence 0⟶OY⟶ν∗(OC1⊕OC2)→ res1−res2 OSpec⁡E⟶0.0\longrightarrow\mathcal O_Y\longrightarrow \nu_*(\mathcal O_{C_1}\oplus\mathcal O_{C_2}) \xrightarrow{\ \mathrm{res}_1-\mathrm{res}_2\ } \mathcal O_{\operatorname{Spec}E}\longrightarrow0. Here the last sheaf is extended by zero along the closed immersion. The cohomology of the finite pushforward equals that on C1⨿C2C_1\amalg C_2 by [StacksAlg] (Tag 089W); the same result and affine acyclicity, [StacksAlg] (Tag 01XB), give vanishing of higher cohomology of the last sheaf. The smooth geometrically connected curve CC is geometrically integral: its geometric local rings are normal domains by Tag 033C, so distinct geometric irreducible components cannot meet, and connectedness leaves just one component. Hence H0(C,OC)=KH^0(C,\mathcal O_C)=K by [StacksAlg] (Tag 04L2, part (4)), the cohomology sequence is 0⟶H0(Y,OY)⟶K2→(a,b)↦a−bE⟶H1(Y,OY)⟶K2⟶0.0\longrightarrow H^0(Y,\mathcal O_Y)\longrightarrow K^2 \xrightarrow{(a,b)\mapsto a-b}E \longrightarrow H^1(Y,\mathcal O_Y)\longrightarrow K^2 \longrightarrow0. It follows that H0(Y,OY)=KH^0(Y,\mathcal O_Y)=K and dim⁡KH1(Y,OY)=d+1\dim_KH^1(Y,\mathcal O_Y)=d+1. Every geometric irreducible component has genus one, so there is no rational tail or bridge. Hence YY is stable by [StacksAlg] (Tag 0E75). All maps and both component labels are defined over KK, which proves the equivariance assertion. ◻

Lemma 4.37.2. NAG-002205 (Normalizing a nodal chart over a valuation ring). Let OO be a valuation ring with fraction field FF, and let a∈Oa\in O. Put Ra=O[u,v]/(uv−a)R_a=O[u,v]/(uv-a). The integral closure of RaR_a in the normalization of (Ra)F(R_a)_F is Ra′={Ra,a≠0,O[u]⊕O[v],a=0.R_a'= \begin{cases} R_a,&a\ne0,\\ O[u]\oplus O[v],&a=0. \end{cases} The ring Ra′R_a' is normal, finite and finitely presented over RaR_a, and flat and finitely presented over OO. Its geometric fibres over OO are smooth or nodal curves.

Proof. A valuation ring is integrally closed in its fraction field: if x∉Ox\notin O satisfied a monic equation over OO, divide it by its leading power of xx; all other terms lie in the maximal ideal, whereas the constant term is one. For a≠0a\ne 0, substitution v=a/uv=a/u identifies RaR_a with the Laurent polynomials ∑ncnun\sum_nc_nu^n satisfying cn∈O(n≥0),cn∈a−nO(n<0).c_n\in O\quad(n\geq0),\qquad c_n\in a^{-n}O\quad(n<0). Indeed the monomials 1,u,u2,…,v,v2,…1,u,u^2,\ldots,v,v^2,\ldots span the quotient and their Laurent images are linearly independent over OO. Inside F[u,u−1]F[u,u^{-1}] we therefore have Ra=O[u,u−1]∩O[v,v−1],v=a/u.R_a=O[u,u^{-1}]\cap O[v,v^{-1}],\qquad v=a/u. Both rings on the right are normal domains by [StacksAlg] (Tags 030A, 00GY); their intersection is integrally closed in the common fraction field F(u)F(u), since a monic equation with coefficients in the intersection is a monic equation over each ring. Hence RaR_a is normal and is already integrally closed in its generic ring F[u,u−1]F[u,u^{-1}]. The displayed monomial basis also proves its OO-flatness.

For a=0a=0, identify R0R_0 with the pairs in O[u]⊕O[v]O[u]\oplus O[v] having the same constant term. The larger ring is generated as an R0R_0-module by the two idempotents (1,0),(0,1)(1,0),(0,1), and has the finite presentation R0′=R0[e]/(e2−e,(1−e)u,ev),e=(1,0).R_0'=R_0[e]/\bigl(e^2-e,(1-e)u,ev\bigr), \qquad e=(1,0). It is normal, as a product of two normal polynomial rings. If an element of F[u]⊕F[v]F[u]\oplus F[v] is integral over R0R_0, each projection is integral over the corresponding polynomial ring over OO, so belongs to that ring. This proves the asserted integral-closure identity. It is free as an OO-module and has an explicit finite presentation over OO. Finally the fibres of RaR_a are uv=a‾uv=\overline a, and those of R0′R_0' are two disjoint affine lines. These equations prove the geometric-fibre assertion. ◻

Proposition 4.37.3. NAG-002206 (Separating the generic components of a nodal model). Let FF be a complete nontrivially valued field, put O=F∘O=F^\circ, and let Y/O\mathcal Y/O be a projective flat finitely presented nodal curve family. Suppose that all special nodes admit the split pointed charts of Theorem 4.15.3 over OO. Suppose that the normalization of the generic fibre is a disjoint union of finitely many smooth proper geometrically connected curves C1,…,CmC_1,\ldots,C_m. The relative normalization of Y\mathcal Y in ∐iCi\coprod_i C_i is a finite projective morphism N→Y\mathcal N\to\mathcal Y, and N=∐i=1mCi,(Ci)F=Ci.\mathcal N=\coprod_{i=1}^m\mathcal C_i, \qquad (\mathcal C_i)_F=C_i. Each Ci/O\mathcal C_i/O is normal, projective, flat and of finite presentation, with reduced nodal special fibre and split special nodes. Every semilinear isometric automorphism of Y\mathcal Y which preserves the labels of the generic normalization induces automorphisms of all Ci\mathcal C_i, compatibly with composition.

Proof. Choose π∈O\pi\in O with 0<∣π∣<10<|\pi|<1. The rank-one property gives O[1/π]=FO[1/\pi]=F, so the generic fibre is the quasi-compact open D(π)D(\pi). The normalization map ∐iCi→YF\coprod_i C_i\to\mathcal Y_F is finite by [StacksAlg] (Tag 0BXR, part (1)), since YF\mathcal Y_F is of finite type over FF. Followed by this open immersion, it is quasi-compact and quasi-separated. Its relative normalization is therefore defined by the integral-closure algebra and has the universal property of [StacksAlg] (Tag 035I).

Relative normalization commutes with smooth, hence with étale, base change by [StacksAlg] (Tag 03GV). On an actual pointed nodal chart Y←E→Spec⁡Ra\mathcal Y\leftarrow E\to\operatorname{Spec}R_a, the generic normalization is the pullback of that of (Ra)F(R_a)_F; one can apply the same result over FF, or identify the two normal branches directly. Thus the pullback of N\mathcal N to EE is E×Spec⁡RaSpec⁡Ra′.E\times_{\operatorname{Spec}R_a} \operatorname{Spec}R_a'. Lemma 4.37.2 gives its finite presentation and finiteness over EE. At smooth points the normalization changes nothing: a smooth algebra AA over the normal ring OO is normal by [StacksAlg] (Tag 033C), and its OO-flatness embeds A[1/π]A[1/\pi] in its total quotient ring. The images of the finitely many nodal charts and the smooth locus cover Y\mathcal Y: they cover the closed fibre, and any omitted generic point would have a closed specialization by properness, whereas this covering is open. Choose finitely many affine charts in the smooth locus to obtain a quasi-compact surjective étale cover. Finiteness now descends by [StacksAlg] (Tag 02LA).

On the induced étale cover of N\mathcal N, each ring is smooth over one of the normal rings Ra′R_a' and is therefore normal by Tag 033C. Normality descends along faithfully flat local maps by [StacksAlg] (Tag 033G). Flatness and local finite presentation over OO descend on this source cover by [StacksAlg] (Tags 036K, 036N). Since N→Y\mathcal N\to\mathcal Y is finite and Y/O\mathcal Y/O is projective, N/O\mathcal N/O is quasi-compact and separated; it is consequently of finite presentation. The same charts show that its geometric fibres are reduced and nodal of pure dimension one. Above an a=0a=0 node the two branches become smooth; above an a≠0a\ne 0 node the normalization is locally the identity, so the remaining special nodes are split. By [StacksAlg] (Tags 0B3I, 0C4P), finiteness and the projectivity of Y/O\mathcal Y/O also make N/O\mathcal N/O projective.

It remains to separate the labels. On any affine open Spec⁡A⊂N\operatorname{Spec}A\subset\mathcal N, OO-flatness makes π\pi a nonzerodivisor, so A[1/π]A[1/\pi] embeds in the total quotient ring of AA. Each labelled generic idempotent satisfies ei2−ei=0e_i^2-e_i=0. A normal ring is integrally closed in its total quotient ring by [StacksAlg] (Tag 034M); hence ei∈Ae_i\in A. The extensions agree on overlaps because multiplication by π\pi is injective on every stalk. They remain orthogonal and sum to one, and therefore give the stated clopen decomposition. Each piece inherits all the preceding geometric properties. Finally the relative integral-closure construction is functorial under semilinear isomorphisms; the generic label idempotents are preserved, so their unique extensions are preserved as well. This proves equivariance and the composition law. ◻

Lemma 4.37.4. NAG-002207 (Stable extension of a nodal generic curve). Let KK be a complete nontrivially valued characteristic-zero field, and let Y/KY/K be a stable curve of genus g≥2g\geq 2. There is a finite Galois extension L/KL/K, with the original absolute value uniquely extended, and a projective stable model Y/L∘\mathcal Y/L^\circ carrying the canonical semilinear Gal⁡(L/K)\operatorname{Gal}(L/K)-action on its generic fibre. After enlarging LL finitely and Galois, all special nodes admit split pointed charts over L∘L^\circ.

Proof. Apply Lemma 4.14.1 to the KK-object YY of the proper stack M‾g,K∘\overline{\mathcal M}_{g,K^\circ}, using [StacksAlg] (Tags 0E9C, 0CL6). This gives a stable family after a finite extension. The projectivity argument in the proof of Theorem 4.14.2 applies without change: Tag 0E74 makes the dualizing line bundle relatively ample over an open neighborhood of the closed point of the local base; that neighborhood is the whole base. Tags 0D32 and 0B5N then give a projective scheme. This argument uses stability of the fibres, not smoothness of the generic fibre.

Take a finite normal closure as in Corollary 4.14.3; characteristic zero makes it Galois, and uniqueness of the absolute-value extension preserves the specified valuation. Apply Lemma 4.14.4 to the semilinear twists of the model. Its statement already permits nodal generic fibres. The twist construction and uniqueness argument of Corollary 4.14.5 therefore supply the group action and its composition law. Finally use Theorem 4.18.2 and take a further normal closure. Stability persists under this base change; the same uniqueness argument supplies the action of the enlarged group. Parameters a=0a=0 are permitted in the pointed charts. ◻

Theorem 4.37.5. NAG-002208 (Equivariant marked genus-one models). Let KK be a complete nontrivially valued characteristic-zero field, let C/KC/K be a smooth projective geometrically connected curve of genus one, and let DD be a finite reduced closed subscheme of CC, possibly empty. There is a finite Galois extension L/KL/K and a projective split nodal model C/L∘\mathcal C/L^\circ of CLC_L with a semilinear Gal⁡(L/K)\operatorname{Gal}(L/K)-action extending the generic action, such that the points of DLD_L are rational and have pairwise distinct smooth specializations. Consequently (C∖D)an(C\setminus D)^{\mathrm{an}} has a jointly continuous strong deformation retraction onto a finite graph with finitely many infinite rays, obtained from the marked model skeleton by finite Galois descent with the original absolute value.

Proof. Form the stable double Y/KY/K of Lemma 4.37.1 and apply Lemma 4.37.4. Its generic normalization is CL⨿CLC_L\amalg C_L, with two Galois-invariant labels. Proposition 4.37.3 therefore supplies two projective split nodal models over L∘L^\circ, each preserved by the canonical semilinear Galois action. Choose the first labelled model. Its generic fibre is the original CLC_L with its specified action.

Enlarge L/KL/K finitely and Galois so that all points of DLD_L are rational. Such an extension exists because the finite residue extensions of closed points are separable in characteristic zero. Base change of a split nodal family remains split nodal; its projectivity, flatness and finite presentation persist. The previously constructed action extends to the larger group through the restriction to the earlier normal extension and the action on coefficients. Properness extends each rational marked point uniquely to a section, as in Theorem 4.34.3. That theorem applies to this projective separated model and the Galois-invariant set of marked sections, and yields the required equivariant model with distinct smooth marked specializations.

Apply Theorem 4.32.2, using its characteristic-zero, arbitrary-residue-field alternative. The entire homotopy is equivariant, so it descends by Corollary 4.9.3. For the noncompact punctured space, Theorem 4.9.2 gives the open quotient property after product with [0,1][0,1]. The finite graph quotient, after subdivision at any inverted-edge midpoint, is again a finite graph with finitely many rays. All field extensions carry the unique extending absolute value throughout. ◻

4.38 Marked models over an algebraically closed valued field

Lemma 4.38.1. NAG-002209 (Doubling at a rational point in any characteristic). Let FF be a field of any characteristic and let C/FC/F be a smooth projective geometrically connected genus-one curve with a specified point P∈C(F)P\in C(F). Gluing two labelled copies of CC at PP gives a stable curve of genus two whose normalization is C⨿CC\amalg C.

Proof. Use the pushout and integral-closure construction in the proof of Lemma 4.37.1, with E=FE=F and d=1d=1. That proof used characteristic zero only to ensure separability of the chosen closed-point residue field. Here the residue extension is the identity. The same fibre-product local ring, finite normalization map, exact cohomology sequence, and no-rational-tail-or-bridge criterion therefore apply over FF and after every field extension. The computed genus is d+1=2d+1=2. ◻

Theorem 4.38.2. NAG-002210 (Split marked models and skeleta in every genus). Let KK be algebraically closed, complete and nontrivially non-Archimedean valued, of any characteristic. Let C/KC/K be a smooth projective connected curve, and let D⊂CD\subset C be a finite reduced closed subscheme, possibly empty. There is a projective flat finitely presented split nodal model C/K∘\mathcal C/K^\circ of CC in which the marked sections have pairwise distinct smooth specializations. Its formal completion gives a finite metric skeleton, with one infinite ray for each point of DD, and a jointly continuous strong deformation retraction of (C∖D)an(C\setminus D)^{\mathrm{an}} onto that skeleton.

Proof. Every closed point of CC is KK-rational by [StacksAlg] (Tag 00FV). If g(C)=0g(C)=0, choose one such point PP. It is an effective Cartier divisor of degree one by [StacksAlg] (Tag 0B8Y). The projective-line characterization [StacksAlg] (Tag 0C6U, implication (7) to (1)) gives C≃PK1C\simeq\mathbf P^1_K. Take the smooth projective model PK∘1\mathbf P^1_{K^\circ}.

If g(C)≥2g(C)\geq 2, Theorem 4.14.2 gives a projective stable model after a finite extension of KK. Since KK is algebraically closed, that extension is KK itself. If g(C)=1g(C)=1, choose a rational point and form the genus-two stable double of Lemma 4.38.1. Apply Lemma 4.14.1 to its object of the proper stable-curve stack. Again its finite extension is KK itself. The projectivity subargument of Theorem 4.14.2, as isolated in Lemma 4.37.4, depends only on stability and the local valuation base. It gives a projective stable model Y/K∘\mathcal Y/K^\circ of this nodal double.

In the positive-genus cases, apply Theorem 4.18.2 to the finitely many special nodes. Its finite extension is still KK, so all nodes already have the required rational split pointed charts over K∘K^\circ. In genus one, Proposition 4.37.3, which is valid in every characteristic, separates the two labelled generic components of Y\mathcal Y. Choose either resulting projective split nodal model of CC. This finishes model existence in all three genus cases.

Properness extends the finitely many rational points of DD to sections. Apply Theorem 4.34.3 with the trivial group to make their specializations smooth and pairwise distinct. Finally the residue field kk of KK is algebraically closed: lift any monic polynomial over kk to a monic polynomial over K∘K^\circ and take a root in KK; a root of norm greater than one is excluded by strict domination of the leading term, so reduction gives a root of the original polynomial. Thus every smooth closed residual point has residue field kk. The perfect-residue alternative of Theorem 4.32.2 supplies the stated skeleton and the entire jointly continuous homotopy. ◻

4.39 Pointed formal charts and the ends of tubes

Let KK be complete and nontrivially valued, with its given absolute value. Write O=K∘O=K^\circ, m=K∘∘\mathfrak m=K^{\circ\circ} and k=O/mk=O/\mathfrak m, and choose π∈O\pi\in O with 0<∣π∣<10<|\pi|<1. All formal schemes in this section are admissible and topologically finitely presented over OO. They carry the π\pi-adic topology. For a special-fibre point qq, write ]q[]q[ for its entire inverse image under specialization.

Proposition 4.39.1. NAG-002220 (Pointed formal étale maps). Let f:U→Xf:\mathfrak U\to\mathfrak X be an adic, topologically finitely presented, formally étale morphism. Suppose that q′∈Us(k)q'\in\mathfrak U_s(k) maps to q∈Xs(k)q\in\mathfrak X_s(k). Then the generic-fibre morphism restricts to an analytic isomorphism ]q′[U →∼ ]q[X.]q'[_{\mathfrak U}\ \xrightarrow{\sim}\ ]q[_{\mathfrak X}.

Proof. A formal open containing a special point contains its whole tube, by Proposition 3.17.2. Choose affine formal neighbourhoods of qq and q′q' so that the latter maps into the former. Write them as Spf⁡A0\operatorname{Spf}A_0 and Spf⁡B0\operatorname{Spf}B_0. Lemma 5.23.1 gives a square Jacobian presentation B0≃A0[Z1,…,Zm,Q]/(F1,…,Fm,Qdet⁡J−1)^,J=(∂Fi∂Zj),B_0\simeq \widehat{A_0[Z_1,\ldots,Z_m,Q]/(F_1,\ldots,F_m,Q\det J-1)}, \qquad J=\left(\frac{\partial F_i}{\partial Z_j}\right), with Fi∈A0[Z1,…,Zm]F_i\in A_0[Z_1,\ldots,Z_m]. The lemma’s presentation statement applies to every admissible base A0A_0. The completion and admissible-presentation results 3.14.8 and 3.14.7 identify this with the corresponding separated restricted-series quotient. Its coefficients have norm at most one in the generic quotient norm.

Choose z0i∈Oz_{0i}\in O lifting Zi(q′)∈kZ_i(q')\in k. Reduction gives F‾(z‾0)(q)=0\overline F(\overline z_0)(q)=0 and det⁡J‾(z‾0)(q)≠0\det\overline J(\overline z_0)(q)\ne0. Thus Corollary 4.11.2 applies to the entire tube. Its inverse is the unique analytic solution with ∣Zi−z0i∣<1|Z_i-z_{0i}|<1; it identifies the analytic structure as well as the points. The open restrictions made above retain both tubes, so this is the claimed map. ◻

Lemma 4.39.2. NAG-002221 (The boundary of a formal closed-point tube). Let Xs\mathfrak X_s be a reduced pure one-dimensional scheme and suppose X=XηX=\mathfrak X_\eta is Hausdorff. For a closed point q∈Xsq\in\mathfrak X_s, put Dq=]q[D_q=]q[ and Cq=Dq‾C_q=\overline{D_q} in XX. Then DqD_q is open, CqC_q is compact, and Cq∖Dq ⊆ {vη:q∈{η}‾, η a component generic point of Xs},C_q\setminus D_q\ \subseteq\ \{v_\eta:q\in\overline{\{\eta\}},\ \eta \text{ a component generic point of }\mathfrak X_s\}, where vηv_\eta is the unique component point of Proposition 4.12.2.

Proof. Choose an affine formal open V\mathfrak V containing qq. Its generic fibre VV is compact, contains the entire tube, and is closed in Hausdorff XX. Consequently CqC_q is compact and lies in VV. On each affine formal chart containing qq, the ideal of qq in the special fibre has finitely many generators. Integral lifts f1,…,ftf_1,\ldots,f_t express its tube by ∣fj∣<1|f_j|<1 for all jj, by Proposition 3.20.1. The same proposition and the affinoid net in Proposition 3.17.2 show that these relatively open pieces define an open subset of XX.

For x∈Cqx\in C_q, let p\mathfrak p be its specialization prime in Vs\mathfrak V_s, and let nq\mathfrak n_q be the prime of qq. If p⊈nq\mathfrak p\not\subseteq\mathfrak n_q, choose f‾∈p∖nq\overline f\in\mathfrak p\setminus\mathfrak n_q and an integral lift ff. Then ∣f(x)∣<1|f(x)|<1, while ∣f(y)∣=1|f(y)|=1 for every y∈Dqy\in D_q. The open condition ∣f∣<1|f|<1 in VV contradicts x∈Dq‾x\in\overline{D_q}. Hence p⊆nq\mathfrak p\subseteq\mathfrak n_q. The proper generalizations of a closed point of a pure one-dimensional Noetherian scheme are component generic points. Proposition 4.12.2 identifies each of their analytic fibres with its single point vηv_\eta. This gives the assertion. ◻

Proposition 4.39.3. NAG-002222 (Formal smooth and nodal tube closures). Let Xs\mathfrak X_s be reduced and pure one-dimensional, and suppose X=XηX=\mathfrak X_\eta is Hausdorff. Let q∈Xs(k)q\in\mathfrak X_s(k) admit a pointed diagram (X,q) ⟵ (U,u) ⟶ (S,o)(\mathfrak X,q)\ \longleftarrow\ (\mathfrak U,u) \ \longrightarrow\ (\mathfrak S,o) of adic, topologically finitely presented, formally étale maps, with U\mathfrak U affine and uu rational over kk. Assume S\mathfrak S is either Spf⁡O⟨T⟩,o=(T‾),orSpf⁡O⟨S,T⟩/(ST−a),o=(S‾,T‾),0<∣a∣<1.\begin{aligned} &\operatorname{Spf}O\langle T\rangle, &&o=(\overline T),\\ \text{or}\quad &\operatorname{Spf}O\langle S,T\rangle/(ST-a), &&o=(\overline S,\overline T),\quad 0<|a|<1. \end{aligned} In the smooth case ]q[]q[ is an open unit disc. Its closure is the disc together with the component point above the unique residual component through qq, and is its one-point compactification. In the nodal case ]q[]q[ is the annulus ∣a∣<∣S∣<1|a|<|S|<1. Its two radial ends attach to the component points of the corresponding two branches through qq. Its closure is obtained by adjoining these ends; the two component points may coincide. Every neighbourhood of an attached endpoint in this closure contains the full corresponding radial tail.

Proof. Reduction of the displayed Jacobian presentations modulo m\mathfrak m gives ordinary étale maps on special fibres. Proposition 4.39.1, applied to both arrows, identifies their entire selected tubes. On the standard chart, Proposition 3.20.1 computes them as ∣T∣<1|T|<1 and ∣a∣<∣S∣<1|a|<|S|<1, respectively.

First work in the compact Hausdorff generic fibre of U\mathfrak U. Its special fibre is smooth or nodal near uu. Shrinking about uu preserves the tube and permits us to assume it is reduced and pure one-dimensional. In the smooth case exactly one component passes through uu: the completion of the local special-fibre ring is k[[T]]k[[T]] by [StacksAlg] (Tag 0AGX), and the ring injects into this domain by [StacksAlg] (Tag 00MC). In the nodal case Lemma 4.20.2 gives exactly two components through uu, corresponding to T=0T=0 and S=0S=0. Lemma 4.39.2 restricts the boundary to their component points.

The open unit disc is not compact, since ∣T∣|T| has supremum one and never takes that value. Thus its compact closure must add the unique possible boundary point. A compact Hausdorff space obtained from an open locally compact space by adding one point is its one-point compactification. Its endpoint neighbourhoods contain the complements of the closed discs ∣T∣≤r|T|\le r, r<1r<1: every compact subset of the open disc has max⁡∣T∣<1\max|T|<1.

For the nodal chart, write vTv_T for the component point of the branch T=0T=0, and vSv_S for that of S=0S=0. Their reductions give ∣S(vT)∣=1,∣T(vS)∣=1,∣S(vS)∣=∣a∣.|S(v_T)|=1,\qquad |T(v_S)|=1,\qquad |S(v_S)|=|a|. A net of points of the tube with ∣S∣→1|S|\to1 has a cluster point in its compact closure. Continuity of ∣S∣|S| and the boundary inclusion force every cluster point to be vTv_T. It therefore converges to vTv_T. The same argument treats ∣S∣→∣a∣|S|\to|a| and vSv_S. Applying this to Gauss points proves that both endpoints occur. If an open neighbourhood of vTv_T contained no full outer tail, its compact complement would contain a net with ∣S∣→1|S|\to1, a contradiction. The inner tail is treated in the same way. Equivalently, the complement has maximum ∣S∣<1|S|<1 or minimum ∣S∣>∣a∣|S|>|a|.

An étale map of special-fibre curves sends component generic points to component generic points, by the going-down argument in Lemma 4.20.2. All components through qq occur this way: the local map at uu over qq is faithfully flat, and going down, [StacksAlg] (Tag 00HS), lifts each minimal prime below the prime of uu. Functoriality of specialization and uniqueness of component points identify the analytic images of the endpoints. The image of the compact tube closure in Hausdorff XX is closed and contains ]q[]q[. Conversely every point in that image is in ]q[‾\overline{]q[}, by continuity and density of the tube in its closure. The map is an isomorphism on the open tube, and an endpoint cannot map into it because its specialization is generic. Thus the only possible identification is between the two nodal endpoints. The endpoint-tail assertions descend through this compact-to-Hausdorff quotient. ◻

Definition 4.39.4. NAG-002223 (Semistable formal curve models with pointed charts). Suppose KK is algebraically closed and C/KC/K is a smooth connected projective curve. A point of CanC^{\mathrm{an}} is of type II when its completed residue field has residual transcendence degree one and relative rational value-group rank zero. A pointedly semistable formal model of CanC^{\mathrm{an}} is a quasi-compact admissible formal OO-scheme with an identification of its generic fibre with CanC^{\mathrm{an}}, reduced pure one-dimensional special fibre, and a diagram as in Proposition 4.39.3 at each closed special-fibre point. Morphisms induce the identity on CanC^{\mathrm{an}}. Its vertex set consists of its component points vηv_\eta.

Corollary 4.39.5. NAG-002224 (The specialization partition of a formal curve). For a pointedly semistable formal model, its vertex set is a finite semistable set of type-II points. Its complement is the disjoint union of the smooth open-disc tubes and finitely many nodal open-annulus tubes. These are exactly its connected components. The boundary of each tube is exactly the set of component points of the residual components passing through its special point.

Proof. The vertex set is finite by Proposition 4.12.2. Proposition 4.24.2, applied on a formal affine neighbourhood of each component generic point, gives residual transcendence degree one and relative rational value-group rank zero. Thus each vertex is of type II under the intrinsic convention just stated. Every point of the finite-type special-fibre curve is either closed or a component generic point. The tubes therefore give the asserted disjoint partition, and Proposition 4.39.3 identifies their analytic structures and boundaries. They are open in CanC^{\mathrm{an}} and clopen in the complement of the vertex set. Each is connected, so they are precisely the connected components. The nodal closed points are finite. Put D=XsD=\mathfrak X_s. It is quasi-compact and locally of finite type over kk, hence Noetherian by a finite affine cover and [StacksAlg] (Tag 00FN). For a point qq of DD, put e(q)=dim⁡κ(q)((ΩD/k)q⊗OD,qκ(q)).e(q)=\dim_{\kappa(q)} \bigl((\Omega_{D/k})_q\otimes_{\mathcal O_{D,q}}\kappa(q)\bigr). On an affine chart a finite presentation of the coordinate algebra gives a finite presentation of ΩD/k\Omega_{D/k} by differentiating the defining relations. With n≥1n\geq 1 generators of the module, the locus e(q)≥2e(q)\geq 2 is cut out by the (n−1)(n-1)-minors of the relation matrix. For n=1n=1 the size-zero minor is 11, and for n=0n=0 the locus is empty. Thus these tests define a closed subset N⊂DN\subset D.

Reduce a supplied pointed diagram modulo m=K∘∘\mathfrak m=K^{\circ\circ}. On the affine opens of this diagram both residual maps are of finite presentation and retain the unique square-zero lifting property. Indeed the targets of the lifting test over kk are discrete and annihilated by m\mathfrak m, so a lift for the formal map factors through the special fibre. The two residual maps are therefore étale by [StacksAlg] (Tag 00UR). For either residual affine map A→BA\to B, differentials over kk pull back. To see this, for a BB-module MM and a kk-derivation d:A→Md:A\to M, give B⊕MB\oplus M, with square-zero summand MM, the AA-algebra structure a↦(a∣B,d(a))a\mapsto(a|_B,d(a)). The identity map to the quotient BB has a unique AA-algebra lift from BB, whose second coordinate is the unique extending kk-derivation. This gives ΩB/k=B⊗AΩA/k\Omega_{B/k}=B\otimes_A\Omega_{A/k}. Dimensions of residue fibres are therefore unchanged along either map.

On k[T]k[T] the relative differential module has one free generator. On k[S,T]/(ST)k[S,T]/(ST) the differential presentation has two generators and the single relation T dS+S dTT\,dS+S\,dT, so its rank-at-least-two locus is the origin. The pointed diagrams thus identify the closed points of NN with the nodal closed points. At a point uu over a nodal closed qq, the ideal (S,T)(S,T) generates the maximal ideal of OUs,u\mathcal O_{\mathfrak U_s,u}, by [StacksAlg] (Tag 02GU(1) and (6)) for the right residual map. Consequently the inverse image of the rank locus in Spec⁡OUs,u\operatorname{Spec}\mathcal O_{\mathfrak U_s,u} consists only of its closed point. The left residual map OD,q→OUs,u\mathcal O_{D,q}\to\mathcal O_{\mathfrak U_s,u} is a flat local map by [StacksAlg] (Tag 02GU(1) and (2)). Its map on spectra is surjective by [StacksAlg] (Tag 00HQ(3) and (2)), since the closed point is hit. Pullback of the residue-rank test now shows that qq has no proper generalization in NN.

Every point of the closed Noetherian subspace NN specializes to a closed point there: choose a minimal nonempty closed subset of its closure and a point of that subset. As NN is closed in DD, such a point is also closed in DD. It follows that every point of NN is closed. Each irreducible component then consists of its single generic point, and there are finitely many irreducible components. Thus only finitely many nodal annular tubes occur. ◻

4.40 Bounded sections and specialization

Let KK be a complete nontrivially real-valued non-Archimedean field, with its specified absolute value. Write O=K∘O=K^\circ, m=K∘∘\mathfrak m=K^{\circ\circ}, and k=O/mk=O/\mathfrak m, and choose π∈O\pi\in O with 0<∣π∣<10<|\pi|<1. All formal schemes below are π\pi-adic. An admissible formal algebra is understood to be π\pi-torsion-free and topologically finitely presented over OO. For a formal open U⊂X\mathfrak U\subset\mathfrak X, put DU=sp⁡−1(U)D_{\mathfrak U}=\operatorname{sp}^{-1}(\mathfrak U), with its analytic GG-domain structure.

Proposition 4.40.1. NAG-002226 (Recovery from bounded analytic sections). Let X\mathfrak X be an admissible formal OO-scheme with reduced special fibre and generic fibre XX. The generic-fibre map identifies Γ(U,OX)={f∈Γ(DU,OX):∣f(x)∣≤1 for all x∈DU}\Gamma(\mathfrak U,\mathcal O_{\mathfrak X}) =\{f\in\Gamma(D_{\mathfrak U},\mathcal O_X): |f(x)|\leq1\text{ for all }x\in D_{\mathfrak U}\} for every formal open U\mathfrak U. These identifications commute with restriction. For each n≥0n\geq0, the ideal πnΓ(U,OX)\pi^n\Gamma(\mathfrak U,\mathcal O_{\mathfrak X}) corresponds to the same condition with bound ∣π∣n|\pi|^n.

Proof. On U=Spf⁡A0\mathfrak U=\operatorname{Spf}A_0, put A=A0[1/π]A=A_0[1/\pi]. Lemma 4.12.1 gives A0=A∘A_0=A^\circ and identifies its quotient norm with the spectral norm. Corollary 1.20.3 then gives A0={a∈A:max⁡x∈M(A)∣a(x)∣≤1}.A_0=\{a\in A:\max_{x\in\mathcal M(A)}|a(x)|\leq1\}. The map A0→AA_0\to A is injective by π\pi-torsion-freeness.

For general U\mathfrak U, cover it by formal affines Ui\mathfrak U_i. Proposition 3.17.2 makes their generic fibres an admissible analytic GG-cover of DUD_{\mathfrak U}. A bounded analytic section restricts to an integral section on each Ui\mathfrak U_i by the affine equality. On each affine subopen of Ui∩Uj\mathfrak U_i\cap\mathfrak U_j, the two integral sections agree after inverting π\pi, hence agree before inverting it. They therefore glue in OX\mathcal O_{\mathfrak X}. Conversely, the restrictions of an integral section have norm at most one and their generic domains cover DUD_{\mathfrak U}. This proves the equality and its restriction compatibility. Applying it to f/πnf/\pi^n proves the last assertion. ◻

Corollary 4.40.2. NAG-002227 (A model determined by its specialization map). Let X1\mathfrak X_1 and X2\mathfrak X_2 be admissible formal models of the same analytic space XX, with reduced special fibres. Suppose there is a homeomorphism h:∣X1,s∣→∣X2,s∣h:|\mathfrak X_{1,s}|\to|\mathfrak X_{2,s}| such that h∘sp⁡1=sp⁡2h\circ\operatorname{sp}_1=\operatorname{sp}_2. There is a unique formal isomorphism inducing hh and the identity on XX.

Proof. The underlying formal topology equals the special-fibre topology: on an affine chart, rank one gives πA0=mA0\sqrt{\pi A_0}=\sqrt{\mathfrak m A_0}, as in Proposition 3.17.2. Corresponding formal opens under hh have the same generic domain. Proposition 4.40.1 identifies their rings of sections, restrictions, and πn\pi^n ideals. These identifications give an isomorphism of locally ringed spaces and recover the local adic topologies. Its generic fibre is the identity. Uniqueness follows from the faithfulness in Proposition 3.17.2. ◻

Proposition 4.40.3. NAG-002228 (Uniqueness from vertices and actual tubes). Let XX be Hausdorff, and let X1,X2\mathfrak X_1,\mathfrak X_2 be admissible formal models of XX. Suppose their special fibres are reduced pure curves of finite type over kk. Suppose also that each closed special point has a connected open tube, and that the boundary of this tube consists exactly of the component vertices of the components through that point. If the two models have the same set V⊂XV\subset X of component vertices, there is a unique formal isomorphism between them inducing the identity on XX.

Proof. Proposition 4.12.2 identifies the generic special points with VV, with singleton specialization fibres. All remaining special points are closed. Their tubes partition X∖VX\setminus V into connected open subsets, each also closed in X∖VX\setminus V because its complement is a union of the other tubes. Thus the closed special points in both models are labelled by the same connected components UU of X∖VX\setminus V.

A point labelled UU lies on the special component labelled vv if and only if v∈U‾∖Uv\in\overline U\setminus U. This recovers all generalizations between special points. The topology of a finite-type pure curve is determined by these incidences: every closed subset is a union of some of its finitely many components and finitely many closed points on the remaining components. Indeed, a proper closed subset of an irreducible Noetherian curve has dimension zero and is finite. Hence the pointwise identification just constructed is a homeomorphism. It intertwines specialization on VV and on every UU. Corollary 4.40.2 applies. ◻

Lemma 4.40.4. NAG-002229 (Vertices under a formal domination). Let X1,X2\mathfrak X_1,\mathfrak X_2 be admissible formal models of an analytic space XX, with reduced pure one-dimensional special fibres of finite type over kk. If a formal morphism f:X1→X2f:\mathfrak X_1\to\mathfrak X_2 induces the identity on XX, then V(X2)⊂V(X1).V(\mathfrak X_2)\subset V(\mathfrak X_1).

Proof. Let v∈V(X2)v\in V(\mathfrak X_2), with η=sp⁡2(v)\eta=\operatorname{sp}_2(v) a component generic point. Put p=sp⁡1(v)p=\operatorname{sp}_1(v). Functoriality of specialization gives fs(p)=ηf_s(p)=\eta. If pp were closed, its residue field would be finite over kk by [StacksAlg] (Tag 00FV), whereas the residue field of η\eta has transcendence degree one by [StacksAlg] (Tag 0A21(6)). The induced injective kk-homomorphism κ(η)→κ(p)\kappa(\eta)\to\kappa(p) is impossible. Thus pp is a component generic point, and the singleton-fibre assertion of Proposition 4.12.2 says that its component vertex is precisely vv. ◻

4.41 The finite forests between prescribed vertices

For the rest of this section and the next, assume that KK is algebraically closed. Let C/KC/K be a smooth connected projective curve and put X=CanX=C^{\mathrm{an}}. A split nodal algebraic model means a projective, flat, finitely presented OO-model of CC with reduced nodal special fibre and the split pointed algebraic charts of Theorem 4.15.3. The smoothing parameters at its nodes are nonzero. Theorems 4.13.2 and 4.23.3 identify its completed generic fibre with XX and supply its embedded finite graph and retraction. The space XX is compact Hausdorff by Proposition 2.33.30.

Lemma 4.41.1. NAG-002231 (Connectedness and the genus of the special fibre). The space XX is connected. For every split nodal algebraic model C\mathcal C of CC, its special fibre is connected and has arithmetic genus g(C)g(C).

Proof. Smoothness over the normal field KK makes CC normal by [StacksAlg] (Tag 033C). Its irreducible components cannot meet, so connectedness makes it integral. Its ring of global functions is a finite field extension of the algebraically closed field KK by [StacksAlg] (Tag 04L2(1)). Together with Corollary 2.42.9, this gives H0(XG,OX)=H0(C,OC)=K.H^0(X_G,\mathcal O_X)=H^0(C,\mathcal O_C)=K. A nontrivial clopen decomposition of XX would give a nonconstant idempotent by gluing the functions zero and one on its two members. The open cover is an admissible analytic cover, so this contradicts the displayed equality. Hence XX is connected.

Its continuous retraction onto Σ(C)\Sigma(\mathcal C) has connected image. The vertices and edges of this graph are the special components and their nodes, with their actual incidences. Thus the dual graph is connected, and so is the union of the special components. The fields KK and kk are algebraically closed; these are the two geometric fibres of the rank-one valuation base. They are reduced, connected curves. By [StacksAlg] (Tag 0E6L), R1f∗OCR^1f_*\mathcal O_{\mathcal C} is finite locally free and commutes with base change. Its rank is constant on Spec⁡O\operatorname{Spec}O, and its generic rank is g(C)g(C). This proves the genus assertion. ◻

Lemma 4.41.2. NAG-002232 (Finite forests outside a semistable vertex set). Let C\mathcal C be a split nodal algebraic model with vertex set WW, skeleton Σ\Sigma, and retraction r:X→Σr:X\to\Sigma. Let V⊂WV\subset W be a semistable vertex set, and let UU be a connected component of X∖VX\setminus V meeting WW. Then S=Σ∩U=r(U)S=\Sigma\cap U=r(U) is a nonempty finite tree with its boundary endpoints removed. It has one boundary end if UU is an open ball and two if UU is an open annulus. Each end has a unique limit in VV; the two annular ends are counted separately when their limits coincide. Every special component indexed by W∩UW\cap U is a smooth Pk1\mathbf P^1_k.

Proof. We first record the topology of the complement of Σ\Sigma. Its components are open balls BB, and B‾=B∪{r(B)}.\overline B=B\cup\{r(B)\}. For a smooth residual tube this is Proposition 4.22.2. Inside a nodal tube the annular Gauss retraction leaves residue-disc components. In an annular coordinate each such disc is ∣T−a∣<∣a∣|T-a|<|a|, attached at ζ0,∣a∣\zeta_{0,|a|}. Its closure in a compact closed subannulus is the disc together with this Gauss point, by the Gauss seminorm formula and polynomial factorization. This subannulus is compact and hence closed in XX, so the same is its closure in XX.

The retraction stays in UU. The set UU is open in XX and closed in X∖VX\setminus V, so its boundary lies in VV. If y∈Uy\in U retracted to v∈Vv\in V, then yy would lie in a smooth residual WW-tube attached to vv: residue discs inside a nodal tube retract to interior edge points, not vertices. That smooth tube is open and closed in X∖VX\setminus V and therefore would equal UU. It contains no vertex of WW, a contradiction. Consequently r(U)r(U) avoids VV. It is connected and contains every w∈W∩Uw\in W\cap U, so it lies in UU. Since rr fixes Σ\Sigma, this proves r(U)=Σ∩Ur(U)=\Sigma\cap U. It is a connected part of a finite graph obtained by deleting vertices in VV.

There are no circuits. Use an actual ball or annulus coordinate on UU. Theorem 4.1.1 describes every type II or III point as a Gauss point ζa,t\zeta_{a,t}. At a type II point, removal partitions the analytic projective line into the inward residue discs and the outward disc. To check exhaustiveness, if ∣T−a∣(x)=t|T-a|(x)=t and x≠ζa,tx\ne\zeta_{a,t}, there is bb with ∣b−a∣≤t|b-a|\leq t and ∣T−b∣(x)<t|T-b|(x)<t. Otherwise every linear factor would have its Gauss value at xx, and factorization over KK would give x=ζa,tx=\zeta_{a,t}. At a type III point, the inward and outward pieces give the analogous two-sided separation. These are disjoint open sets and distinguish the two sides of a radial interval.

The Gauss interval between two points of UU lies in UU. In a ball its joining radius is still smaller than the outer radius. In an annulus the value of ∣T∣|T| on the interval is between the two endpoint values, or is their common value. Every interior point of this interval separates its endpoints. Therefore every path between the endpoints contains the interval. An embedded compact arc must equal the interval, since an interval contained in an arc with the same endpoints is the whole arc. This also agrees with the explicit interval calculation of Lemma 4.35.1. A circuit in SS would give two different arcs between two of its Gauss points. Thus SS is a tree.

Radial ends have limits in VV. For a ball U={∣T∣<R}U=\{|T|<R\}, the connected outer tails Et={t<∣T∣<R}E_t=\{t<|T|<R\} have nested nonempty compact connected closures in XX. Their intersection is nonempty and connected. It misses UU and is contained in its finite boundary VV, hence is a singleton {vU}\{v_U\}. Every neighborhood of vUv_U contains a whole tail: otherwise the nested compact sets outside that neighborhood would have a common point. The closed disc ∣T∣≤t|T|\leq t is compact in UU and closed in XX. It follows that every boundary point lies in every E‾t\overline E_t, so ∂XU={vU}\partial_XU=\{v_U\}. For an annulus, the same argument applies to each of its two radial ends. A compact middle subannulus shows that its boundary consists of the two resulting limits v−,v+v_-,v_+; they may be equal.

Each radial end occurs exactly once in the finite tree. Choose a compact disc or middle annulus containing the finite set W∩UW\cap U. Every graph end of SS leaves all compact subsets of UU, because its endpoint in the compact graph Σ\Sigma lies in VV. Its final portion therefore approaches one coordinate end. The cut-point separation just proved forces this portion to contain a sufficiently short radial Gauss tail. Two different graph ends cannot approach the same coordinate end, since they would share a nonempty Gauss interval.

Suppose a coordinate end were unrepresented. Some sufficiently far tail EE would then miss SS. Indeed, if all tails met SS, one of its finitely many edges would have points escaping toward that end; compactness of the closed edge would produce its endpoint in VV and a graph end approaching the chosen coordinate end. The connected set EE lies in one ball component BB of X∖ΣX\setminus\Sigma. Because BB meets UU and avoids VV, we have B⊂UB\subset U. Also r(B)∈r(U)⊂Ur(B)\in r(U)\subset U, so the compact closure B∪{r(B)}B\cup\{r(B)\} misses VV. All shorter tails lie in BB, contrary to the nonempty limiting set in VV proved above. Every coordinate end is therefore represented. This gives exactly one or two ends.

The unwanted components are smooth rational curves. For w∈W∩Uw\in W\cap U, translation and scaling identify its type II Gauss field with the unit Gauss field. Its residue field is k(t)k(t) by Proposition 3.8.1. Proposition 4.24.2 identifies this with the function field of the special component DwD_w. The normalization Dwν→DwD_w^\nu\to D_w is finite by [StacksAlg] (Tag 0BXR). It is therefore proper, and DwνD_w^\nu is proper over kk by [StacksAlg] (Tags 01WM, 01W3). The normal proper curve equivalence [StacksAlg] (Tag 0BY1) now identifies DwνD_w^\nu with Pk1\mathbf P^1_k. A self-node of DwD_w would give a loop based at ww in Σ\Sigma. Its interior is connected, avoids VV, and meets the open set UU near ww, so the entire loop lies in UU. This contradicts acyclicity. Thus DwD_w has no self-node. At a node shared with another component, its own branch is smooth, by the pointed branch calculation in Lemma 4.20.2. Hence DwD_w itself is smooth and equals its normalization, as asserted. ◻

4.42 Exact contraction of auxiliary components

Lemma 4.42.1. NAG-002234 (Pinching smooth sections over a local ring). Let RR be a local ring. Let AA and BB be flat finitely presented RR-algebras with augmentations to RR. Their fibre product is a flat finitely presented RR-algebra and commutes with arbitrary base change. If AA and BB are affine neighborhoods of sections in the smooth loci of separated relative curves, they can be shrunk around those sections so that A×RBA\times_R B is étale over R[u,v]/(uv)R[u,v]/(uv).

Consequently, pinching smooth sections of two proper flat finitely presented nodal RR-curves produces a proper flat finitely presented nodal RR-curve, compatibly with arbitrary base change.

Proof. Translate finite algebra generators by their augmentation values and write A=R[x1,…,xm]/IA=R[x_1,\ldots,x_m]/I and B=R[y1,…,yn]/JB=R[y_1,\ldots,y_n]/J. The relation ideals are finitely generated and have zero constant terms. There is an isomorphism A×RB=R[x1,…,xm,y1,…,yn]/(I,J,(xiyj)i,j).A\times_R B =R[x_1,\ldots,x_m,y_1,\ldots,y_n]/(I,J,(x_i y_j)_{i,j}). Indeed, after removing cross products every polynomial is represented as c+f(x)+g(y)c+f(x)+g(y) modulo I,JI,J, with f,gf,g of zero constant term; its image is zero precisely when c=0c=0, f∈If\in I, and g∈Jg\in J. As an RR-module this ring is R⊕ker⁡(A→R)⊕ker⁡(B→R)R\oplus\ker(A\to R)\oplus\ker(B\to R), hence is flat. The split exact sequence 0⟶A×RB⟶A⊕B→ϵA−ϵBR⟶00\longrightarrow A\times_R B\longrightarrow A\oplus B \xrightarrow{\epsilon_A-\epsilon_B}R\longrightarrow0 proves arbitrary-base-change compatibility. The ring A⊕BA\oplus B is generated as an A×RBA\times_R B-module by (1,0)(1,0) and (0,1)(0,1), and the corresponding finite morphism is surjective.

For the local chart assertion, choose actual étale coordinates at the special points of the sections using [StacksAlg] (Tag 054L), and subtract their section values. The chosen affine neighborhoods contain the entire sections, since an open subset of a local spectrum containing its closed point is the whole spectrum. The zero fibre of an étale coordinate map is étale and separated over RR; its section is open and closed by [StacksAlg] (Tag 024T). Remove its other components and take a principal affine neighborhood of the section’s special point. We obtain R[u]⟶A eˊtale,A/uA=R,R[v]⟶B eˊtale,B/vB=R.R[u]\longrightarrow A\text{ \textup{\'etale}},\quad A/uA=R, \qquad R[v]\longrightarrow B\text{ \textup{\'etale}},\quad B/vB=R. The algebra D=A⊗RBD=A\otimes_R B is étale, hence flat, over R[u,v]R[u,v]. Tensoring 0⟶R[u,v]/(uv)⟶R[u,v]/(v)⊕R[u,v]/(u)⟶R[u,v]/(u,v)⟶00\longrightarrow R[u,v]/(uv) \longrightarrow R[u,v]/(v)\oplus R[u,v]/(u) \longrightarrow R[u,v]/(u,v)\longrightarrow0 with DD identifies its kernel with A×RBA\times_R B. Thus A×RB=D/(uv)A\times_R B=D/(uv), étale over R[u,v]/(uv)R[u,v]/(uv).

The scheme pushout along the two closed sections exists by [StacksAlg] (Tag 0E25). Its affine-neighborhood hypothesis holds because every fibre of a section map has at most one point. Near the joined section its rings are precisely the fibre products above; elsewhere the old curves are unchanged. These charts prove flatness, local finite presentation, and nodality. The disjoint union of the original curves maps finitely and surjectively to the pushout. The pushout is separated by [StacksAlg] (Tag 0E26) and universally closed by [StacksAlg] (Tag 03GN), hence proper. Quasi-compactness and separatedness promote local finite presentation to finite presentation. The same charts and their localization maps commute with arbitrary base change, proving the last assertion. ◻

Lemma 4.42.2. NAG-002235 (Protection by elliptic tails). Let C\mathcal C be a split nodal algebraic model with vertex set WW, and let V⊂WV\subset W be a semistable vertex set of XX. There is a proper flat finitely presented prestable family Y0/O\mathcal Y_0/O obtained by attaching three smooth elliptic tails at distinct smooth points on every component indexed by VV. Its genus is g(C)+3∣V∣g(C)+3|V|, its generic fibre is stable, and its stabilization c:Y0→Yc:\mathcal Y_0\to\mathcal Y contracts precisely the original special components indexed by W∖VW\setminus V. The stable family Y\mathcal Y is projective over OO.

Proof. First choose a smooth projective elliptic OO-curve E\mathcal E with a smooth section. If char⁡k≠2\operatorname{char}k\ne2, take the projective completion of y2=x(x−1)(x−λ),λ‾∉{0,1}.y^2=x(x-1)(x-\lambda), \qquad \overline\lambda\notin\{0,1\}. Its fibres are smooth by the distinct roots and the unit 22. If char⁡k=2\operatorname{char}k=2, take y2+y=x3y^2+y=x^3. The special fibre has yy-derivative one. In mixed characteristic a singular affine point would have x=0x=0 and y=−1/2y=-1/2, which does not satisfy the equation; in equal characteristic the derivative is again one. The point at infinity is smooth in both equations. Lemma 4.36.2 supplies flatness, finite presentation, and projectivity. Each fibre is a smooth plane cubic, so its global functions are the constants and its genus is one by [StacksAlg] (Tag 0BYD). Use the infinity section.

Choose three distinct smooth kk-points on each special component indexed by VV, outside its nodes. Such points exist because a nonempty smooth open of a curve over the algebraically closed field kk has infinitely many closed points: its closed points are dense and rational by [StacksAlg] (Tag 00FV), and finitely many closed points cannot be dense in a curve. By Corollary 4.17.2, each chosen tube contains a KK-point. The valuative criterion [StacksAlg] (Tag 0A40) extends this point uniquely to an OO-section; its specialization is the chosen point. These sections are disjoint and lie in the relative smooth locus: an intersection, or a closed complement of the smooth locus meeting a section, would also meet its special point. Attach a separate copy of E\mathcal E at each section by Lemma 4.42.1.

Both geometric fibres of C\mathcal C are connected of genus g(C)g(C) by Lemma 4.41.1. On either fibre, over its algebraically closed ground field FF, one elliptic attachment gives 0⟶OA∪E⟶OA⊕OE⟶Op⟶0.0\longrightarrow\mathcal O_{A\cup E} \longrightarrow\mathcal O_A\oplus\mathcal O_E \longrightarrow\mathcal O_p\longrightarrow0. The final map on global sections is the surjective difference F⊕F→FF\oplus F\to F. Its cohomology sequence gives H0(A∪E,O)=FH^0(A\cup E,\mathcal O)=F and H1(A∪E,O)=H1(A,O)⊕H1(E,O)H^1(A\cup E,\mathcal O)=H^1(A,\mathcal O)\oplus H^1(E,\mathcal O). Thus the fibres remain connected and their genus is g(C)+3∣V∣g(C)+3|V|. The set VV is nonempty: otherwise connected compact XX would itself be an open ball or annulus. Hence this genus is at least three. The connected nodal-fibre criterion [StacksAlg] (Tags 0E6L, 0E6T) makes Y0\mathcal Y_0 prestable.

Apply stabilization [StacksAlg] (Tag 0E8A). Its fibre maps contract rational tails and then rational bridges as described in [StacksAlg] (Tag 0E7Q). In a component UU of X∖VX\setminus V containing an unwanted vertex, Lemma 4.41.2 gives a finite tree of smooth rational components with one or two labelled boundary ends. In a one-ended tree, successively remove leaves away from the boundary end. These are rational tails and remove the whole tree. In a two-ended tree, first remove all leaves away from the unique path joining its ends; then contract the remaining rational bridges. The two boundary ends remain distinct branch labels if they attach to the same component in VV, and contracting the path creates a self-node on that component. Every unwanted vertex is thereby removed. At each stage every VV-component retains its three auxiliary elliptic attachments, and every elliptic component retains its one attachment. The remaining components are stable. The uniqueness of stabilization proves that exactly these components survive.

In the generic fibre the original smooth curve has at least three attachments and each elliptic tail has one, so this fibre is already stable. Its stabilization is therefore its identity. Finally [StacksAlg] (Tag 0E74) makes the relative dualizing line bundle of Y\mathcal Y ample over a neighborhood of the closed point of Spec⁡O\operatorname{Spec}O, hence over the whole base. By [StacksAlg] (Tags 0D32, 0B5N), Y\mathcal Y is a projective scheme over this local base. ◻

Lemma 4.42.3. NAG-002236 (Lifting to a generic normalization). Let OO be a rank-one valuation domain with fraction field KK. Let N→Y\mathcal N\to\mathcal Y be a finite relative normalization in a specified finite normalization of YK\mathcal Y_K. Let TT be a normal integral flat OO-scheme. A morphism f:T→Yf:T\to\mathcal Y with a specified lift on TKT_K has a unique lift T→NT\to\mathcal N. If the generic lift lands in one clopen generic-labelled piece of N\mathcal N, then the whole lift lands in that piece. If T/OT/O is proper and N/O\mathcal N/O is separated, the lift is proper.

Proof. Choose π\pi with O[1/π]=KO[1/\pi]=K. Flatness makes TKT_K dense in TT. For Q=Spec⁡R⊂YQ=\operatorname{Spec}R\subset\mathcal Y, write NQ=Spec⁡B\mathcal N_Q=\operatorname{Spec}B. The algebra BB is integral over RR. The specified generic lift pulls each b∈Bb\in B back to a rational function in K(T)K(T). Its monic equation over RR pulls back to a monic equation over every local ring of f−1(Q)f^{-1}(Q). Those rings are integrally closed in K(T)K(T), so the rational function is regular at every point of f−1(Q)f^{-1}(Q). Its local representatives glue, yielding an RR-homomorphism B→Γ(f−1(Q),OT)B\to\Gamma(f^{-1}(Q),\mathcal O_T). These maps agree on overlaps because their generic restrictions agree and TKT_K is dense. They define the lift, and the same argument proves uniqueness. The inverse image of any other clopen labelled piece is open and misses the dense generic fibre, so is empty. For properness, factor the lift through its closed graph in T×ONT\times_O\mathcal N and the proper projection to N\mathcal N; this is also [StacksAlg] (Tag 01W6). ◻

Theorem 4.42.4. NAG-002237 (Exact realization by a split nodal model). Let KK be algebraically closed, complete and nontrivially valued, with its original absolute value. Let C/KC/K be a smooth connected projective curve and let VV be a semistable vertex set of CanC^{\mathrm{an}}. There is a split nodal algebraic model CV/O\mathcal C_V/O whose component vertices are exactly the original points of VV. More precisely, from every split nodal algebraic model C0\mathcal C_0 with vertex set containing VV, there is a proper morphism C0→CV\mathcal C_0\to\mathcal C_V inducing the identity on CC.

Proof. Theorem 4.38.2, with empty marking, supplies a split nodal model, and Theorem 4.27.4 supplies one whose vertex set WW contains VV. Start with any such C0\mathcal C_0. Apply Lemma 4.42.2, obtaining C0→Y0→Y\mathcal C_0\to\mathcal Y_0\to\mathcal Y. The generic normalization of the stable curve YK\mathcal Y_K is the disjoint union of CC and the labelled auxiliary elliptic curves. The special nodes of Y\mathcal Y have the required split pointed charts: apply Theorem 4.18.2; its finite extension of the algebraically closed field KK is KK. Proposition 4.37.3 therefore gives a finite relative normalization N=CV⨿∐jEj⟶Y.\mathcal N=\mathcal C_V\amalg\coprod_j\mathcal E_j \longrightarrow\mathcal Y. The CC-labelled piece CV\mathcal C_V is projective, flat and finitely presented with split reduced nodal special fibre.

The source C0\mathcal C_0 is integral and normal. Flatness embeds the ring of each nonempty affine open into its generic ring, the ring of a nonempty open of the integral curve CC; this proves integrality. Smooth points are normal over the normal valuation ring by [StacksAlg] (Tag 033C). At a special node, the smoothing parameter is nonzero, so Lemma 4.37.2 gives normality of its standard ring O[u,v]/(uv−a)O[u,v]/(uv-a). Normality ascends along the pointed étale chart and descends to the model local ring by [StacksAlg] (Tags 033C, 033G). These loci cover C0\mathcal C_0.

The generic restriction of C0→Y\mathcal C_0\to\mathcal Y has its specified lift to the CC-labelled generic normalization, since the generic stabilization is the identity. Lemma 4.42.3 gives a proper morphism q:C0→CVq:\mathcal C_0\to\mathcal C_V inducing the identity on CC.

By Lemma 4.42.2, the original surviving special components are precisely those indexed by VV. At their generic points Y\mathcal Y is relatively smooth, so relative normalization is the identity there. The local normalization formula of Lemma 4.37.2 introduces no extra vertical components: at a≠0a\ne0 it is the identity, and at a=0a=0 it separates the two branches. Each auxiliary elliptic model is normal by smoothness over the normal base and is integral by flatness and its integral generic fibre. The same normalization-lift lemma puts its surviving special component in its own generic-labelled piece Ej\mathcal E_j. Thus these auxiliary components belong to none of the other pieces, and CV,s\mathcal C_{V,s} has exactly the surviving original components.

For v∈Vv\in V, let ηv\eta_v be its original component generic point. The point qs(ηv)q_s(\eta_v) is the generic point of its surviving component in CV,s\mathcal C_{V,s}. Functoriality of specialization for the completed morphism gives sp⁡CV(v)=qs(ηv).\operatorname{sp}_{\mathcal C_V}(v)=q_s(\eta_v). The singleton-fibre theorem 4.12.2 identifies the new component vertex with the original point vv. Thus the vertex set is exactly VV as a subset of CanC^{\mathrm{an}}. ◻

Theorem 4.42.5. NAG-002238 (The vertex correspondence for algebraic completions). Let KK and CC be as in Theorem 4.42.4. Consider the formal completions of split nodal algebraic models of CC, with their specified identifications of the generic fibre with X=CanX=C^{\mathrm{an}}. Their isomorphism classes over XX correspond bijectively to the semistable vertex sets of XX. For two such models CW\mathfrak C_W and CV\mathfrak C_V, there is a formal morphism CW→CV\mathfrak C_W\to\mathfrak C_V inducing the identity on XX if and only if V⊂WV\subset W; when it exists, it is unique. The graph attached to CV\mathfrak C_V is Σ(X,V)\Sigma(X,V) with its original logarithmic edge lengths.

Proof. The actual smooth and nodal tube theorems and Theorem 4.23.3 show that every such model has a semistable vertex set and the asserted graph. Theorem 4.42.4 gives every semistable vertex set. Its uniqueness as a formal model follows from Proposition 4.40.3, using Proposition 4.22.2 and Proposition 4.20.3 for the exact boundaries.

If V⊂WV\subset W, apply the contraction in Theorem 4.42.4 to an algebraic model underlying CW\mathfrak C_W and complete the resulting morphism. The target has vertices VV, so the preceding uniqueness identifies it with CV\mathfrak C_V over XX. Conversely, Lemma 4.40.4 gives the required inclusion for any such formal morphism. Its uniqueness is generic-fibre faithfulness. Finally, the nodal tubes are exactly the annular components of X∖VX\setminus V; their Gauss intervals give Σ(X,V)\Sigma(X,V), and Corollary 4.16.2 identifies their lengths with their logarithmic moduli. ◻

Corollary 4.42.6. NAG-002239 (The correspondence for pointed semistable formal models). Let KK be algebraically closed, complete and nontrivially valued, and let C/KC/K be a smooth connected projective curve. The vertex-set functor is an equivalence between the pointedly semistable formal models of Definition 4.39.4 and the partially ordered set of semistable vertex sets, with a unique arrow W→VW\to V when V⊂WV\subset W. Every such formal model is isomorphic over CanC^{\mathrm{an}} to the completion of a split nodal algebraic model. Its skeleton and logarithmic edge lengths are those of its semistable vertex set.

Proof. By Corollary 4.39.5, a pointedly semistable formal model has a semistable vertex set VV and the exact tube boundaries required in Proposition 4.40.3. Theorem 4.42.4 constructs a split nodal algebraic model CV\mathcal C_V with that vertex set. Its completion belongs to the pointed formal class. Indeed, complete the pointed algebraic smooth and nodal charts π\pi-adically. Their rings are admissible by Lemma 4.13.1. The completed maps are topologically finitely presented. They are formally étale because, on a discrete target annihilated by a power of π\pi, maps from a completed ring factor through its corresponding algebraic quotient, where the algebraic étale lifting property [StacksAlg] (Tag 00UR) gives the unique lift.

Proposition 4.40.3 now identifies the given formal model with CV^\widehat{\mathcal C_V} over CanC^{\mathrm{an}}. The existence and uniqueness of arrows are consequently Theorem 4.42.5; alternatively the necessary vertex inclusion is already Lemma 4.40.4. The skeletal statement follows from the same identification and the final assertion of that theorem. ◻

4.43 Finite unramified coefficients and integral sections

Let KK be a complete nontrivially real-valued field, with its given absolute value. Put O=K∘O=K^\circ, m=K∘∘\mathfrak m=K^{\circ\circ} and k=K~k=\widetilde K, and fix π∈m∖{0}\pi\in\mathfrak m\setminus\{0\}. A finite unramified extension in this section is the lift L/KL/K of a finite separable extension l/kl/k constructed in Lemma 4.15.2. Thus [L:K]=[l:k][L:K]=[l:k] and ∣L×∣=∣K×∣|L^\times|=|K^\times|.

Lemma 4.43.1. NAG-002262 (Integral Galois descent for unramified coefficients). Suppose l/kl/k is finite Galois, take its Galois lift L/KL/K, and write O′=L∘O'=L^\circ and G=Gal⁡(L/K)G=\operatorname{Gal}(L/K). Then O′/OO'/O is finite free and étale, and O′⊗OO′→∼∏g∈GO′,a⊗b⟼(a g(b))g.O'\otimes_O O'\xrightarrow{\sim}\prod_{g\in G}O', \qquad a\otimes b\longmapsto(a\,g(b))_g. For every OO-module MM, the coefficient inclusion identifies M=(M⊗OO′)G.M=(M\otimes_O O')^G. If A0A_0 is an admissible formal OO-algebra, its coefficient extension is the complete admissible algebra A0⊗OO′A_0\otimes_O O'. Its generic algebra is A0[1/π]⊗^KLA_0[1/\pi]\widehat\otimes_K L.

Proof. Write O′=O[α]=O[T]/(P)O'=O[\alpha]=O[T]/(P) as in Lemma 4.22.1, with 1,α,…,αd−11,\alpha,\ldots,\alpha^{d-1} an orthonormal OO-basis. The polynomial PP is monic, and P′(α)P'(\alpha) has nonzero reduction. For a square-zero ideal, a lift bb of a root is corrected uniquely to a root by b−P(b)/P′(b)b-P(b)/P'(b); successive square-zero quotients give the same assertion for a nilpotent ideal. Thus the finite presentation is formally étale, hence étale by [StacksAlg] (Tag 00UR). Lemma 4.18.1 gives all its roots g(α)g(\alpha) in O′O', with pairwise different residues. Evaluation on these roots is an isomorphism: its matrix on 1,T,…,Td−11,T,\ldots,T^{d-1} has Vandermonde determinant ∏g<h(h(α)−g(α))∈(O′)×\prod_{g<h}(h(\alpha)-g(\alpha))\in(O')^\times. This proves the tensor-product assertion, as well as faithful flatness.

The fixed module assertion holds without dividing by dd. Here is an explicit descent argument. Let z=∑imi⊗bi∈M⊗OO′z=\sum_i m_i\otimes b_i\in M\otimes_O O' be invariant. Under the displayed product isomorphism, the two elements ∑imi⊗bi⊗1,∑imi⊗1⊗biin M⊗OO′⊗OO′\sum_i m_i\otimes b_i\otimes1, \qquad \sum_i m_i\otimes1\otimes b_i \quad\text{in }M\otimes_O O'\otimes_O O' have equal gg-coordinates for every gg, so are equal. Apply the OO-linear coefficient map λ:O′→O\lambda:O'\to O which takes the coefficient of 11 in the indicated basis; it satisfies λ(1)=1\lambda(1)=1. Applying λ\lambda on the first coefficient factor gives z=(∑iλ(bi)mi)⊗1z=(\sum_i\lambda(b_i)m_i)\otimes1. Injectivity of m↦m⊗1m\mapsto m\otimes1 follows from the same coefficient map. This proves the equality of modules.

A finite free sum commutes with π\pi-adic completion. The extended ring is π\pi-torsion-free, and its finite presentation follows by extending an admissible presentation of A0A_0. For the generic algebra put A=A0[1/π]A=A_0[1/\pi]. Inverting π\pi identifies (A0⊗OO′)[1/π](A_0\otimes_O O')[1/\pi] with A⊗KLA\otimes_K L by the coefficient maps. In Proposition 2.33.19 take the finite algebra to be LL over KK and the target to be AA. The actual absolute value on LL is its finite-module maximum norm in the displayed basis. The resulting maximum norm on A⊗KLA\otimes_K L and the defining quotient norm of (A0⊗OO′)[1/π](A_0\otimes_O O')[1/\pi] given by Proposition 3.14.7 have the same closed unit ball A0⊗OO′A_0\otimes_O O'. Scaling by powers of π\pi compares the norms in both directions with constant ∣π∣−1|\pi|^{-1}, giving the required bounded isomorphism with A⊗^KLA\widehat\otimes_K L. All these identifications preserve the chosen inclusion of KK and its absolute value. ◻

Proposition 4.43.2. NAG-002263 (Bounded sections after an unramified extension). Keep the Galois extension and notation of Lemma 4.43.1. Let X\mathfrak X be an admissible formal OO-scheme with reduced special fibre, let X′\mathfrak X' be its coefficient extension to O′O', and let X′=XLX'=X_L be the generic fibre. For a formal open U⊂X\mathfrak U\subset\mathfrak X, write U′\mathfrak U' for its inverse image and D,D′D,D' for the corresponding generic domains. Then Γ(U,OX)=Γ(U′,OX′)G={f∈Γ(D′,OX′)G:∣f∣≤1}.\Gamma(\mathfrak U,\mathcal O_{\mathfrak X}) =\Gamma(\mathfrak U',\mathcal O_{\mathfrak X'})^G =\{f\in\Gamma(D',\mathcal O_{X'})^G: |f|\leq1\}. These equalities commute with restriction and identify the πn\pi^n ideals by the bound ∣π∣n|\pi|^n.

Proof. On an affine formal open the first equality is Lemma 4.43.1 with M=A0M=A_0. Its special-fibre extension is (A0/mA0)⊗kl(A_0/\mathfrak mA_0)\otimes_k l, which is reduced by the separable-extension argument in Lemma 4.22.1. Proposition 4.40.1 therefore identifies the integral sections upstairs with the displayed bounded analytic sections. The actual generic domain upstairs is DLD_L by the coefficient presentation and the generic-fibre gluing of Proposition 3.17.2.

For a general formal open, restrict an invariant integral section to the inverse images of a formal affine cover. The affine equalities descend it uniquely on each member. Their restrictions agree because the upstairs restrictions agree and coefficient extension is injective. The descended sections consequently glue. The converse and restriction compatibility follow from the same affine identifications. Apply the equality to f/πnf/\pi^n for the last assertion. In particular this descent recovers the actual integral sheaf and its adic topology. ◻

4.44 Auxiliary multisections and generic normalization

Lemma 4.44.1. NAG-002265 (Smooth finite étale multisections). Let C/O\mathcal C/O be a proper flat finitely presented curve, and let qq be a smooth closed point of Cs\mathcal C_s with l=κ(q)l=\kappa(q) finite separable over kk. There is a closed immersion Spec⁡L∘⟶C\operatorname{Spec}L^\circ\longrightarrow\mathcal C through qq, for a finite unramified lift L/KL/K of l/kl/k. Its image is contained in the relative smooth locus. Distinct choices of the special points give disjoint such multisections.

If Cs\mathcal C_s is nodal, a finite union of its irreducible components can be met by a finite disjoint union of these multisections, avoiding any prescribed finite set of smooth points and all nodes, with at least three geometric intersection points on each geometric component in that union and none on the other components.

Proof. Choose the unramified lift by Lemma 4.15.2. Its valuation ring has the monic presentation O[T]/(P)O[T]/(P) of Lemma 4.22.1, with P′P' a unit at the lifted root. The square-zero correction in the proof of Lemma 4.43.1 does not require a Galois extension: it proves this presentation formally étale. Thus L∘/OL^\circ/O is finite étale by [StacksAlg] (Tag 00UR). After this coefficient extension, the map Spec⁡l→Cs\operatorname{Spec}l\to\mathcal C_s gives a rational smooth point of CL∘,s\mathcal C_{L^\circ,s}. Its open-disc tube contains an LL-point by Corollary 4.17.2. The valuative criterion, [StacksAlg] (Tag 0A40), extends that point to a section over L∘L^\circ, with the given specialization. Compose with the coefficient projection to obtain i:Spec⁡L∘→Ci:\operatorname{Spec}L^\circ\to\mathcal C.

Choose an affine neighborhood Spec⁡A\operatorname{Spec}A of qq. Its inverse image contains the closed point of the local spectrum Spec⁡L∘\operatorname{Spec}L^\circ, hence that whole spectrum. Choose a∈Aa\in A whose image in ll is a primitive generator β\beta over kk. Put d=[l:k]d=[l:k]. The elements 1,i∗(a),…,i∗(a)d−11,i^*(a),\ldots,i^*(a)^{d-1} are an OO-basis of L∘L^\circ: relative to the basis in Lemma 4.22.1, their change-of-basis matrix reduces to the change between two bases of l/kl/k. Its determinant is a unit of OO. Thus A→L∘A\to L^\circ is surjective. The morphism ii is finite, since its graph is a closed subscheme of C×OSpec⁡L∘\mathcal C\times_O\operatorname{Spec}L^\circ and projection from this product is finite. Consequently the closed immersion just proved on the neighborhood of its entire image is a closed immersion into C\mathcal C. The inverse image of the smooth locus is an open neighborhood of the closed point and is therefore the whole source.

Two multisections with different special points cannot meet: their intersection is closed in a scheme finite over OO; if nonempty, its proper image in the local spectrum contains the closed point, contradicting their different specializations.

It remains to choose enough separable points. Choose an algebraic closure of kk by [StacksAlg] (Tag 09GT) and work over the separable closure ksepk^{\mathrm{sep}} of kk inside it. Every nonempty open in the smooth part of a geometric component contains a ksepk^{\mathrm{sep}}-point. Indeed, a smooth coordinate gives an étale map to the affine line ([StacksAlg] (Tag 054L)); its image is open ([StacksAlg] (Tag 03WT)). The infinite field ksepk^{\mathrm{sep}} has a rational point in this image, and the nonempty étale fibre has finite separable residue fields by [StacksAlg] (Tag 02GL), hence rational points. A purely inseparable extension does not split the underlying components: for such a field extension the prime over a given prime is unique, because membership in it is detected after taking sufficiently high pp-powers. Thus this argument applies to each geometric component.

Choose three points on each required component, successively avoiding the finitely many previously chosen Galois orbits and the prescribed exceptional points. Each chosen point is defined over a finite separable extension of kk, since the curve is of finite type. Its image is a closed separable point of the original curve. Take all the finitely many resulting closed points and lift them by the first part. Their geometric specializations have the required incidence and disjointness. ◻

Lemma 4.44.2. NAG-002266 (Pinching along finite étale markings). Let ZZ be finite étale over OO, and let Z↪AZ\hookrightarrow\mathcal A, Z↪BZ\hookrightarrow\mathcal B be closed immersions into the smooth loci of proper flat finitely presented nodal OO-curves. Their pushout is a proper flat finitely presented nodal OO-curve. Its formation commutes with arbitrary base change on OO.

Proof. The pushout exists by [StacksAlg] (Tag 0E25): each fibre of either closed immersion contains at most one point, so the required affine-neighborhood condition holds. On compatible affine neighborhoods its ring is A×CBA\times_C B, where both maps onto the coordinate ring CC of the corresponding open in ZZ are surjective. This ring CC is flat over OO. There is an exact sequence 0⟶A×CB⟶A⊕B→a−bC⟶0.0\longrightarrow A\times_C B\longrightarrow A\oplus B \xrightarrow{a-b}C\longrightarrow0. Let P=A×CBP=A\times_C B and Q=A⊕BQ=A\oplus B, so that 0→P→Q→C→00\to P\to Q\to C\to0 is the displayed sequence of ordinary OO-modules. For an arbitrary OO-algebra TT, choose a surjection from a free OO-module FF onto its underlying module and write RR for the kernel. If a class in P⊗OTP\otimes_O T maps to zero, lift it to pFp_F in P⊗OFP\otimes_O F. The image of pFp_F in Q⊗OFQ\otimes_O F comes from an element qRq_R of Q⊗ORQ\otimes_O R. The image of qRq_R in C⊗ORC\otimes_O R maps to zero in C⊗OFC\otimes_O F, so is zero because CC is flat. Lift qRq_R to P⊗ORP\otimes_O R. These lifts exist because tensoring a module quotient is the quotient of the tensor product by the image of the relations, by the defining balanced maps. Tensoring the inclusion P→QP\to Q with the free module FF is a direct sum of copies of that inclusion. It follows that pFp_F is the image of the lift in P⊗ORP\otimes_O R, so the original class vanishes in P⊗OTP\otimes_O T. The same quotient property identifies the image with the kernel of (A⊗OT)⊕(B⊗OT)→C⊗OT(A\otimes_O T)\oplus(B\otimes_O T)\to C\otimes_O T. The resulting isomorphism with the fibre product is an algebra map by its two coordinate maps. Thus these affine pushouts and their localizations commute with every base change.

Choose a finite Galois extension of kk splitting the finite étale scheme ZsZ_s, and take its unramified lift by Lemma 4.18.1. Its integral coefficient extension is faithfully flat by Lemma 4.43.1. Over its valuation ring O′O', every rational point of ZsZ_s lifts to an open-and-closed section of the finite étale scheme ZO′Z_{O'}, by Lemma 4.15.1. These sections exhaust it: their complement is finite over O′O' and has empty special fibre, hence is empty. The base-changed pushout is therefore the successive pinching of disjoint smooth sections. Lemma 4.42.1 proves it flat, finitely presented and nodal. These properties descend along the finite faithfully flat cover: use [StacksAlg] (Tags 02L0, 0C5E), the latter including flatness and relative dimension one.

Both old curves map by closed immersions into the pushout, as is also clear from the two surjective projections of A×CBA\times_C B. Their disjoint union is a finite surjective cover. The pushout is separated by [StacksAlg] (Tag 0E26) and universally closed over OO by [StacksAlg] (Tag 03GN). Together with the finite presentation just established, these properties give properness. The affine kernel calculation already proves the stated base-change compatibility. ◻

Proposition 4.44.3. NAG-002267 (Generic normalization of a nodal family). Let Y/O\mathcal Y/O be a projective flat finitely presented nodal curve family. Its relative normalization in the normalization of YK\mathcal Y_K is finite over Y\mathcal Y, normal, projective, flat and finitely presented over OO, with reduced nodal geometric fibres. A decomposition of the generic normalization into open-and-closed pieces extends uniquely to such a decomposition of this relative normalization. The construction is functorial for semilinear isometric isomorphisms, including permutations of the generic pieces.

In particular a nodal OO-model with smooth geometrically integral generic fibre is normal and integral.

Proof. The generic normalization is finite by [StacksAlg] (Tag 0BXR). Since O[1/π]=KO[1/\pi]=K, its composite with the generic open immersion is quasi-compact and quasi-separated. The relative integral-closure construction of [StacksAlg] (Tag 035I) therefore defines ν:N→Y\nu:\mathcal N\to\mathcal Y.

Theorem 4.18.2 supplies a finite unramified Galois extension L/KL/K giving split pointed charts at every special node. Its valuation ring is finite étale and faithfully flat over OO by Lemma 4.43.1. Relative normalization commutes with this base change and with the pointed étale charts by [StacksAlg] (Tag 03GV). On a chart Ra=O′[u,v]/(uv−a)R_a=O'[u,v]/(uv-a) its actual algebra is Ra′=Ra(a≠0),R0′=O′[u]⊕O′[v],R_a'=R_a\quad(a\ne0),\qquad R_0'=O'[u]\oplus O'[v], by Lemma 4.37.2. At smooth points it is the identity, since a smooth algebra over the normal valuation ring is normal ([StacksAlg] (Tag 033C)). The finitely many nodal charts and the smooth locus cover the family: an omitted generic point would have a specialization in the proper closed fibre, which these open sets cover. These formulas prove, on an actual étale cover, finiteness over YO′\mathcal Y_{O'}, normality, and flat finite presentation over O′O', with reduced nodal geometric fibres. They require no choice of labels for the generic components.

Finiteness descends by [StacksAlg] (Tag 02LA). Normality descends on local rings by [StacksAlg] (Tag 033G); flatness and local finite presentation descend on the induced source cover by [StacksAlg] (Tags 036K, 036N). The finite morphism to projective Y\mathcal Y gives quasi-compactness and separatedness, hence finite presentation. Nodality descends by [StacksAlg] (Tag 0C5E). Finally finiteness and projectivity of Y/O\mathcal Y/O give projectivity of N/O\mathcal N/O by [StacksAlg] (Tags 0B3I, 0C4P).

On every affine open of N\mathcal N, flatness makes π\pi a nonzerodivisor. A labelled generic idempotent is an element of the total quotient ring satisfying e2−e=0e^2-e=0. Normality and [StacksAlg] (Tag 034M) put it in the original affine ring. These extensions agree on overlaps by π\pi-torsion-freeness; they remain orthogonal and sum to one. This proves the asserted unique decomposition. Integral closure and these idempotents are carried to their counterparts by every indicated isomorphism, proving functoriality.

For the last assertion, a flat model of the integral generic curve is integral: each affine coordinate ring embeds in its generic coordinate ring, a domain. In the preceding local charts its smoothing parameter is nonzero, because the generic fibre is smooth. Thus the relative normalization is locally the identity after the faithfully flat cover. The original model is consequently normal as well. ◻

Lemma 4.44.4. NAG-002268 (Proper contractions as ringed-space quotients). Let q:T→Yq:T\to Y be a proper birational morphism of integral schemes, with YY normal. Then q∗OT=OYq_*\mathcal O_T=\mathcal O_Y. Every morphism h:T→Zh:T\to Z which is constant on the set-theoretic fibres of qq factors uniquely through a morphism Y→ZY\to Z.

Proof. Let U=Spec⁡A⊂YU=\operatorname{Spec}A\subset Y be affine. A section f∈Γ(q−1U,OT)f\in\Gamma(q^{-1}U,\mathcal O_T) is a rational function in the common function field FF. For every valuation ring A⊂R⊂FA\subset R\subset F with fraction field FF, properness extends the generic map to TT to a map Spec⁡R→q−1U\operatorname{Spec}R\to q^{-1}U, by [StacksAlg] (Tag 0A40). Therefore f∈Rf\in R. The normal domain AA is the intersection of these valuation rings by [StacksAlg] (Tag 090P), so f∈Af\in A. The reverse inclusion is pullback. These equalities commute with restriction and prove the sheaf assertion.

The proper map qq is surjective, since its closed image contains the generic point of YY. It is therefore a closed quotient map on underlying spaces. The map hh induces a unique continuous map hˉ:Y→Z\bar h:Y\to Z. For an open U⊂ZU\subset Z, pullback by hh, followed by the sheaf equality, gives OZ(U)⟶OT(q−1hˉ−1U)=OY(hˉ−1U).\mathcal O_Z(U)\longrightarrow \mathcal O_T(q^{-1}\bar h^{-1}U) =\mathcal O_Y(\bar h^{-1}U). These maps respect restrictions. They are local on stalks: choose tt above yy; if the image in OY,y\mathcal O_{Y,y} of an element of mZ,hˉ(y)\mathfrak m_{Z,\bar h(y)} were a unit, its image in OT,t\mathcal O_{T,t} would be a unit, contrary to locality of hh. Hence hˉ\bar h is a morphism of schemes. Surjectivity and the injective sheaf pullback give uniqueness. ◻

4.45 Geometric vertices and marked forests

Lemma 4.45.1. NAG-002270 (An algebraically closed comparison field). There is an isometric extension K⊂ΩK\subset\Omega with Ω\Omega complete and algebraically closed and with Ω~\widetilde\Omega an algebraic closure of kk.

Proof. Choose an algebraic closure K‾\overline K by [StacksAlg] (Tag 09GT). The unique absolute values of its finite subextensions, supplied by Lemma 1.12.1, agree on intersections and give an absolute value on K‾\overline K. Let Ω\Omega be its completion. Completion changes neither the value group nor the residue field: approximate a nonzero element with error smaller than its absolute value, or a unit with error smaller than one, respectively.

We verify algebraic closedness. Let P(T)=Tn+∑i<naiTi∈Ω[T]P(T)=T^n+\sum_{i<n}a_iT^i\in\Omega[T] be monic. Choose monic Pj∈K‾[T]P_j\in\overline K[T] with coefficients converging to those of PP. Their coefficients are uniformly bounded, so all their roots have absolute value at most one fixed R≥1R\geq1: beyond RR, the leading term would strictly dominate all the others. Write ϵj\epsilon_j for the maximum coefficient difference between PjP_j and Pj+1P_{j+1} and choose a root α1\alpha_1 of P1P_1. If αj\alpha_j has been chosen, then ∣Pj+1(αj)∣≤ϵjRn−1.|P_{j+1}(\alpha_j)|\leq\epsilon_j R^{n-1}. Factor Pj+1P_{j+1} over K‾\overline K. At least one of its roots αj+1\alpha_{j+1} satisfies ∣αj−αj+1∣≤(ϵjRn−1)1/n|\alpha_j-\alpha_{j+1}|\leq (\epsilon_jR^{n-1})^{1/n}. These bounds tend to zero, so the ultrametric inequality makes (αj)(\alpha_j) Cauchy. Its limit is a root of PP. Induction on the degree now shows that Ω\Omega is algebraically closed.

The residue field of K‾\overline K is algebraic over kk: each of its elements comes from a finite extension, whose residue degree is finite by [StacksAlg] (Tag 0ASH). It is algebraically closed, since lifting a monic residual polynomial and taking a root in K‾\overline K produces an integral root and hence a residual root. Completion retains that residue field, proving the assertion. ◻

Proposition 4.45.2. NAG-002271 (The arithmetic of a model vertex). Let C/O\mathcal C/O be a projective flat finitely presented nodal model of a smooth projective curve. If vv is the component point belonging to a special component EE, then ∣H(v)×∣=∣K×∣,H(v)~=k(E).|\mathcal H(v)^\times|=|K^\times|, \qquad \widetilde{\mathcal H(v)}=k(E). The field k(E)/kk(E)/k is separably generated of transcendence degree one. For the comparison field Ω\Omega above, the entire inverse image of vv in CΩanC_\Omega^{\mathrm{an}} consists of the component points of the geometric components above EE. In particular it is finite.

Proof. The equalities are Proposition 4.24.2 on a completed affine neighborhood of the generic point of EE. A nodal curve is smooth at its component generic points. A smooth étale coordinate, as in [StacksAlg] (Tags 054L, 02GL), presents each function field as a finite separable extension of k(t)k(t). This proves the separability and transcendence-degree assertions.

Lemma 4.28.1 identifies the actual base-changed model and its specialization map. The special fibre is Cs×kΩ~\mathcal C_s\times_k\widetilde\Omega and remains reduced. Every point over the generic point η\eta of EE is a component generic point. To check this, its fibre ring is k(E)⊗kΩ~k(E)\otimes_k\widetilde\Omega, an integral algebraic extension of the field k(E)k(E) and hence zero-dimensional by [StacksAlg] (Tag 00GS(3)). Any prime properly below a prime over η\eta would also lie over the minimal prime η\eta, contradicting this zero-dimensionality. The points over η\eta are therefore precisely the finitely many geometric component generic points. If an analytic point projects to vv, its specialization lies over η\eta, so it is one of their singleton component fibres by Proposition 4.12.2. Conversely each of those component points projects to vv by functorial specialization and the same singleton-fibre statement. This proves the entire-fibre assertion. ◻

Definition 4.45.3. NAG-002272 (Marked semistable models over the original field). Let C/KC/K be smooth, projective and geometrically connected. A semistable algebraic model of CC means a projective flat finitely presented nodal OO-curve with generic fibre identified with CC and geometrically connected special fibre. Its special fibre is geometrically reduced. A marking is a finite étale OO-scheme DD embedded as a closed subscheme of the relative smooth locus. An empty marking is allowed. Write WW for the component vertex set of the model. A morphism of marked models carries a specified isomorphism of its marking with the target marking, compatible with the curve morphism and with the specified generic identifications.

The geometric marked dual graph has one vertex for each component over k‾\overline k, one edge for each geometric node, and one labelled leg for each point of Dk‾D_{\overline k}. The two branches of a self-node give the two ends of its loop.

Fix a nonempty subset V⊂WV\subset W, and let V‾\overline V denote all geometric vertices over its special components. For a connected component TT of the subgraph induced by the vertices outside V‾\overline V, let b(T)b(T) count its incident edges leading to V‾\overline V, with multiplicity, and let m(T)m(T) count its marked legs.

Definition 4.45.4. NAG-002273 (The marked forest condition). The subset VV satisfies the marked forest condition if every such TT is a tree, all the geometric special components at its vertices are smooth Pk‾1\mathbf P^1_{\overline k}, and b(T)+m(T)≤2.b(T)+m(T)\leq2. Connectedness and V≠∅V\ne\varnothing give b(T)≥1b(T)\geq1. Thus a tree with one attaching branch carries at most one mark, and a tree with two attaching branches carries none. The two attaching branches may end on the same protected component.

Theorem 4.45.5. NAG-002274 (Exact marked contraction over the original field). Let (C0,D)(\mathcal C_0,D) be a marked semistable algebraic model of CC, and let ∅≠V⊂W\varnothing\ne V\subset W satisfy the marked forest condition. There is a marked semistable algebraic model (CV,DV)(\mathcal C_V,D_V) and a proper morphism qV:C0⟶CVq_V:\mathcal C_0\longrightarrow\mathcal C_V inducing the identity on CC, with D→∼DVD\xrightarrow{\sim}D_V and with component vertices exactly the original points of VV in CanC^{\mathrm{an}}.

On a geometric special fibre, the positive-dimensional fibres of qVq_V are exactly the trees TT in Definition 4.45.4. A tree with b(T)=1b(T)=1 maps to a smooth point, and a tree with b(T)=2b(T)=2 maps to a node with the indicated attaching branches. Outside these fibres the special-fibre map is an isomorphism. Moreover (qV)∗OC0=OCV(q_V)_*\mathcal O_{\mathcal C_0}=\mathcal O_{\mathcal C_V}.

Proof. Protect the desired components and markings. Lemma 4.44.1 gives a finite étale multisection AA on the protected special components, disjoint from DD and from the nodes, meeting each protected geometric component at least three times. There exists a smooth projective elliptic curve E/O\mathcal E/O with a section. If char⁡k≠2\operatorname{char}k\ne2, use y2=x(x−1)(x−λ)y^2=x(x-1)(x-\lambda) with λ‾∉{0,1}\overline\lambda\notin\{0,1\}; such a residue exists even when k=F3k=\mathbf F_3. If char⁡k=2\operatorname{char}k=2, use y2+y=x3y^2+y=x^3. The distinct-root and derivative calculations in the proof of Lemma 4.42.2 show smoothness of both fibres and of the infinity section. Lemma 4.36.2 gives flatness, finite presentation and projectivity, and [StacksAlg] (Tag 0BYD) gives geometric genus one.

Put Z=A⨿DZ=A\amalg D. Attach E×OZ\mathcal E\times_O Z to C0\mathcal C_0 along ZZ and its infinity section. Lemma 4.44.2 produces a proper flat finitely presented nodal family Y0/O\mathcal Y_0/O. On either geometric fibre, each attachment adds a smooth elliptic tail at one point. The exact sequence 0⟶OF∪E⟶OF⊕OE⟶Op⟶00\longrightarrow\mathcal O_{F\cup E} \longrightarrow\mathcal O_F\oplus\mathcal O_E \longrightarrow\mathcal O_p\longrightarrow0 has a surjective difference map on constants. Its cohomology therefore preserves geometric connectedness and adds one to the genus. The original fibres have the same genus g(C)g(C) by [StacksAlg] (Tag 0E6L), applied to the proper flat family with geometrically reduced connected fibres. Consequently Y0\mathcal Y_0 is prestable of genus g(C)+deg⁡(Z/O)≥3g(C)+\deg(Z/O)\geq3. Its generic fibre is stable: the original smooth component has at least three attachments, and each elliptic component has one.

Stabilize and separate the generic components. Apply [StacksAlg] (Tag 0E8A) to obtain the stabilization c:Y0→Yc:\mathcal Y_0\to\mathcal Y. Its generic map is an isomorphism. On the geometric special fibre it contracts rational tails and bridges as in [StacksAlg] (Tag 0E7Q). In a removed tree, regard its b(T)b(T) attachments to protected components and its m(T)m(T) elliptic marking tails as distinguished ends. There are one or two such ends. With one end, successively remove leaves away from it. With two ends, first remove the leaves off their unique joining path and then contract the rational bridges on that path. Every removed vertex disappears. Every protected component keeps its three auxiliary elliptic attachments, and every elliptic component keeps its attachment; all these components are stable. Uniqueness of stabilization consequently identifies these as exactly the surviving components. In the two-ended case the ends retain separate branch labels even if they lie on the same protected component.

The stable family Y\mathcal Y is projective over OO: [StacksAlg] (Tag 0E74) makes its dualizing line bundle ample over a neighborhood of the closed point, hence over the entire local base, and [StacksAlg] (Tags 0D32, 0B5N) give projectivity. Apply Proposition 4.44.3 to Y\mathcal Y. Its generic normalization has the distinguished component CC and the elliptic components attached along ZKZ_K. Let CV\mathcal C_V be its open-and-closed CC-labelled piece. It is normal, projective, flat and finitely presented with nodal geometric fibres.

The source C0\mathcal C_0 is normal and integral by the same proposition. The morphism C0→Y0→Y\mathcal C_0\to\mathcal Y_0\to\mathcal Y has its specified generic lift to CC. Lemma 4.42.3 therefore gives an actual proper morphism qV:C0→CVq_V:\mathcal C_0\to\mathcal C_V inducing the identity on CC.

Identify the fibres, marks, and original vertices. The normalization calculation is the identity at smooth points and at nodes with nonzero smoothing parameter; at a node with zero parameter it separates the two generic branches. Hence it introduces no vertical component. Each surviving auxiliary or marking elliptic component belongs to its own generic-normalization piece: apply the same normalization-lift lemma to its smooth normal elliptic model after an unramified coefficient extension splitting ZZ. Thus CV\mathcal C_V contains precisely the protected original components.

The geometric fibre maps of stabilization are compositions of the elementary rational-tail and rational-bridge contractions. They identify an unmarked one-ended tree with its smooth attaching point. A two-ended tree without marks gives a node joining its two original branches. For a one-ended tree with one mark, its second distinguished end is the marking elliptic tail. Stabilization joins that tail to the protected component, and generic normalization separates this horizontal node. The mark therefore lands at the resulting smooth point of CV\mathcal C_V. These descriptions also cover marks on protected components. Distinct marks remain distinct: each removed one-ended tree contains at most one, different such trees have different attaching points, and the original protected marks avoid those nodal attachments. They show that the claimed trees are exactly the positive-dimensional geometric fibres and that the map is an isomorphism away from them. In particular the special fibre of CV\mathcal C_V is geometrically connected.

The map D→CVD\to\mathcal C_V is finite by the closed-graph argument, since D/OD/O is finite. After a finite unramified cover splitting DD, it is the union of the just identified disjoint smooth sections, hence a closed immersion. On an affine open of CV\mathcal C_V, its ring map is therefore surjective after the faithfully flat coefficient extension; the cokernel vanishes already over OO. This proves the closed immersion. Its image is smooth because all its generic and geometric special points are smooth in the nodal family.

At the generic point ηv\eta_v of any protected original component, stabilization is an isomorphism and generic normalization is the identity. Thus qV(ηv)q_V(\eta_v) is the generic point of the corresponding new component. Complete the projective models as in Theorem 4.13.2. Functoriality of specialization in Proposition 3.17.2 gives sp⁡CV(v)=qV(ηv).\operatorname{sp}_{\mathcal C_V}(v)=q_V(\eta_v). Proposition 4.12.2 identifies its singleton analytic fibre with the original point vv. Every surviving component has just been accounted for, so the vertex set is exactly VV. Finally apply Lemma 4.44.4 to the proper birational map qVq_V and its normal target. ◻

Proposition 4.45.6. NAG-002275 (Necessity of the geometric marked forest condition). Let q:(C0,D)→(C1,D1)q:(\mathcal C_0,D)\to(\mathcal C_1,D_1) be a morphism of marked semistable algebraic models inducing the identity on CC and on the generic marking. Then V(C1)⊂V(C0)V(\mathcal C_1)\subset V(\mathcal C_0), and this subset satisfies the marked forest condition in C0\mathcal C_0.

Proof. Complete qq along π\pi. Lemma 4.40.4 gives the vertex inclusion. Extend the original absolute value isometrically to the comparison field Ω\Omega of Lemma 4.45.1. Both base-changed models are actual projective flat nodal Ω∘\Omega^\circ-models of CΩC_\Omega, and their analytic generic fibres are CΩanC_\Omega^{\mathrm{an}} by Lemma 4.28.1. Their nodes have split rational pointed charts: Theorem 4.18.2 applies, and a finite extension of the algebraically closed field Ω\Omega is Ω\Omega itself. Thus the algebraically closed model results apply to these models.

Proposition 4.45.2 identifies their vertex sets with the entire inverse images of the two original vertex sets. Put W‾=V(C0,Ω)\overline W=V(\mathcal C_{0,\Omega}) and V‾=V(C1,Ω)\overline V=V(\mathcal C_{1,\Omega}). For a component UU of CΩan∖V‾C_\Omega^{\mathrm{an}}\setminus\overline V meeting W‾\overline W, Lemma 4.41.2 shows that the source skeleton in UU is a finite tree with one or two boundary ends, and all its component curves are smooth Pk‾1\mathbf P^1_{\overline k}. Its vertices and internal edges are exactly one connected component TT of the induced removed graph. Its boundary ends are exactly the b(T)b(T) attaching branches, counted separately even when they end at the same protected component.

The set UU is the actual tube of a closed point pp of the target. For every removed component with generic point η\eta, functorial specialization gives qs(η)=pq_s(\eta)=p. Since pp is closed, every point of that component, including all its marked specializations, also maps to pp. If b(T)=1b(T)=1, the target point pp is smooth, and the closed immersion D1,k‾D_{1,\overline k} has at most one geometric point over it. Thus m(T)≤1m(T)\leq1. If b(T)=2b(T)=2, then pp is nodal and the marking avoids it, so m(T)=0m(T)=0. This proves the stated condition on every removed geometric tree. ◻

Proposition 4.45.7. NAG-002276 (Algebraic uniqueness of an exact contraction). Let qi:C0→Ciq_i:\mathcal C_0\to\mathcal C_i, i=1,2i=1,2, be morphisms of semistable algebraic models inducing the identity on CC. If the targets have the same component vertices in CanC^{\mathrm{an}}, there is a unique algebraic isomorphism h:C1→C2h:\mathcal C_1\to\mathcal C_2 with q2=h∘q1q_2=h\circ q_1. It induces the identity on CC.

Proof. After extension to Ω\Omega, the target vertex sets are the same by Proposition 4.45.2. The split pointed charts supplied by Theorem 4.18.2 complete to the charts of Definition 4.39.4, as in the proof of Corollary 4.42.6. Corollary 4.39.5 gives the connected open tubes and their exact component-incidence boundaries. Their completed models are isomorphic over CΩanC_\Omega^{\mathrm{an}} by Proposition 4.40.3. Write hsh_s for the resulting special-fibre isomorphism. It intertwines their specialization maps. Functoriality of specialization and its surjectivity on the source, both in Proposition 3.17.2, give hs∘(q1)k‾=(q2)k‾h_s\circ(q_1)_{\overline k} =(q_2)_{\overline k} on the underlying geometric special-fibre points. Consequently (q2)k‾(q_2)_{\overline k} is constant on every set-theoretic fibre of (q1)k‾(q_1)_{\overline k}.

We descend this assertion about actual fibres. Every field extension is faithfully flat: its only base ideals are zero and the whole field, so [StacksAlg] (Tag 00HD) gives flatness, and its nonzero closed fibre gives faithful flatness by [StacksAlg] (Tag 00HP). If t1,t2∈C0,st_1,t_2\in\mathcal C_{0,s} have the same image yy under (q1)s(q_1)_s, the tensor product κ(t1)⊗κ(y)κ(t2)\kappa(t_1)\otimes_{\kappa(y)}\kappa(t_2) is nonzero by the same faithful-flatness criterion. By [StacksAlg] (Tag 00E0(2)), a prime of this tensor product supplies a point of C0,s×C1,sC0,s\mathcal C_{0,s}\times_{\mathcal C_{1,s}}\mathcal C_{0,s} whose two projections are t1,t2t_1,t_2. The extension k‾/k\overline k/k is faithfully flat by the preceding argument. On an affine base change, its tensor functor on modules is the original coefficient tensor functor, so remains faithfully flat. Surjectivity on spectra, [StacksAlg] (Tag 00HQ), lifts the chosen point to the base-changed fibre product. The lifted projections have the same (q1)k‾(q_1)_{\overline k}-image, hence the same (q2)k‾(q_2)_{\overline k}-image. Projecting back gives (q2)s(t1)=(q2)s(t2)(q_2)_s(t_1)=(q_2)_s(t_2). On the generic fibre q1q_1 and q2q_2 are identities. These two fibres exhaust the rank-one base, and their images cannot meet. Thus q2q_2 is constant on every fibre of q1q_1.

The morphisms qiq_i are proper: their graphs are closed in the products over OO, and projection from those products is proper. Their targets are normal and integral by Proposition 4.44.3. Lemma 4.44.4 gives an actual algebraic factorization hh. Interchanging the indices gives its inverse; uniqueness of factorization shows that the composites are the identity. This proves both the assertion and its uniqueness. ◻

Theorem 4.45.8. NAG-002277 (The ordered category of marked contractions). Fix a marked semistable algebraic model (C0,D)(\mathcal C_0,D). The category of its marked semistable algebraic contractions over CC is equivalent to the following partially ordered set: its objects are the nonempty subsets VV of V(C0)V(\mathcal C_0) satisfying the geometric marked forest condition, and there is one arrow V′→VV'\to V precisely when V⊂V′V\subset V'. A contraction here includes its actual morphism from C0\mathcal C_0, and arrows commute with these morphisms.

Proof. Existence is Theorem 4.45.5, necessity is Proposition 4.45.6, and uniqueness for each object is Proposition 4.45.7.

Suppose V⊂V′V\subset V'. On the geometric special fibre, every tree removed for V′V' lies in a tree removed for VV: its vertices are still removed, and its connected edges remain in the induced removed graph. The fibre description in Theorem 4.45.5 therefore shows that qVq_V is constant on every geometric fibre of qV′q_{V'}. The fibre-product argument in the proof of Proposition 4.45.7 descends this to the original special fibre. The generic assertion is immediate. Lemma 4.44.4 gives the unique factorization C0→qV′CV′⟶CV.\mathcal C_0\xrightarrow{q_{V'}} \mathcal C_{V'}\longrightarrow\mathcal C_V. It preserves the markings because its composite with D↪C0D\hookrightarrow\mathcal C_0 does. Conversely, every such arrow gives V⊂V′V\subset V' by Lemma 4.40.4. Composites are forced by uniqueness, giving the claimed equivalence. ◻

Corollary 4.45.9. NAG-002278 (Semilinear group actions on exact contractions). Let a group GG act on (C0,D)(\mathcal C_0,D) by semilinear isometric automorphisms, and let VV be invariant and satisfy the marked forest condition. There is a unique action of GG on (CV,DV)(\mathcal C_V,D_V) making qVq_V equivariant. Its generic action is the prescribed action on CC. Inclusions between invariant admissible vertex sets give equivariant contraction morphisms.

Proof. Put S=Spec⁡OS=\operatorname{Spec}O, X0=C0X_0=\mathcal C_0, Y=CVY=\mathcal C_V, and q=qVq=q_V. Write p0p_0 and pYp_Y for the structure maps. For g∈Gg\in G, let τg\tau_g be its automorphism of SS, so that p0g=τgp0p_0g=\tau_gp_0. Form the actual base change Yg=S×τg,S,pYYY_g=S\times_{\tau_g,S,p_Y}Y with projections pg:Yg→Sp_g:Y_g\to S and jg:Yg→Yj_g:Y_g\to Y. The second projection is an isomorphism of schemes, since τg\tau_g is an isomorphism, and pYjg=τgpgp_Yj_g=\tau_gp_g. The pair (p0,qg)(p_0,qg) defines an OO-morphism hg:X0→Ygh_g:X_0\to Y_g with jghg=qgj_gh_g=qg.

To check constancy on fibres, let σ\sigma be the induced automorphism of kk. Apply [StacksAlg] (Tag 09GV) to k‾\overline k with the two kk-structures given by the inclusion and its composite with σ\sigma. This gives an automorphism of k‾\overline k extending σ\sigma. With this choice, the geometric special fibre of hgh_g is identified, via the corresponding semilinear isomorphism of geometric target fibres, with the composite of the geometric action of gg and the geometric contraction. The geometric action preserves the marked graph and permutes the removed trees because VV is invariant. The geometric fibre description of Theorem 4.45.5 makes this composite constant on every set-theoretic fibre of the geometric contraction, including the singleton fibres. Apply the fibre-product descent argument in the proof of Proposition 4.45.7 to the kk-morphisms qsq_s and (hg)s(h_g)_s. It gives constancy on the actual special fibres of qsq_s. The generic map of qq is an isomorphism. These two fibres exhaust the rank-one base, so hgh_g is constant on all set-theoretic fibres of qq.

The contraction qq is proper birational between integral schemes with normal target. By Lemma 4.44.4, there is a unique hˉg:Y→Yg\bar h_g:Y\to Y_g such that hˉgq=hg\bar h_gq=h_g. The two maps pghˉgp_g\bar h_g and pYp_Y to SS agree after composition with qq. Uniqueness in the same lemma makes them equal, so hˉg\bar h_g is an OO-morphism. Set gV=jghˉgg_V=j_g\bar h_g. Then qg=gVqqg=g_Vq and pYgV=τgpYp_Yg_V=\tau_gp_Y. The generic action is the prescribed one because qq is the specified identification on the generic fibre. Uniqueness of factorization through qq gives (gh)V=gVhV(gh)_V=g_Vh_V and the identity element. In particular (g−1)V(g^{-1})_V is the inverse of gVg_V. Since q∣D:D→DVq|_D:D\to D_V is the specified isomorphism, the equality qg=gVqqg=g_Vq gives the induced action on the marking as a scheme. Finally, for an arrow of Theorem 4.45.8 from Y′=CV′Y'=\mathcal C_{V'} to YY, the two equivariance composites agree after precomposition with qV′:X0→Y′q_{V'}:X_0\to Y'. Uniqueness of factorization through qV′q_{V'} makes them equal. ◻

4.46 Algebraic morphisms determined by original-field vertices

Let KK be complete and nontrivially real-valued, with its specified absolute value. Put O=K∘O=K^\circ, m=K∘∘\mathfrak m=K^{\circ\circ}, and k=K~k=\widetilde K, and fix 0≠π∈m0\ne\pi\in\mathfrak m. Let C/KC/K be a smooth projective geometrically connected curve. Semistable algebraic models and finite étale smooth markings have the meaning of Definition 4.45.3. Every generic-fibre identification in this section is an identification with this fixed curve over the original field.

Lemma 4.46.1. NAG-002310 (The actual closure of the generic graph). Let X1,X2\mathcal X_1,\mathcal X_2 be projective flat finitely presented OO-models of CC. The schematic closure G\mathcal G of the generic diagonal in X1×OX2\mathcal X_1\times_O\mathcal X_2 is integral, projective, flat and finitely presented over OO. Its two projections pi:G⟶Xip_i:\mathcal G\longrightarrow\mathcal X_i are proper and induce the identity on CC. If Xi\mathcal X_i is integral, pip_i is birational and surjective.

For every complete isometric extension L/KL/K, the completion of GL∘\mathcal G_{L^\circ} is an admissible formal model of CLanC_L^{\mathrm{an}}. Its specialization is surjective onto every point of its special fibre and satisfies sp⁡Xi,L∘=(pi)L~∘sp⁡GL∘.\operatorname{sp}_{\mathcal X_{i,L^\circ}} =(p_i)_{\widetilde L}\circ \operatorname{sp}_{\mathcal G_{L^\circ}}.

Proof. On an affine open Spec⁡B\operatorname{Spec}B of the product, let I⊂B[1/π]I\subset B[1/\pi] be the ideal of the generic diagonal. Its closure has ring B/(I∩B).B/(I\cap B). These ideals commute with localization and hence define a closed subscheme of the product. A nonempty such affine closure has coordinate ring contained in the coordinate ring of a nonempty open of the integral curve CC. It is therefore a domain and has no OO-torsion. The generic diagonal is dense, so these affines give an integral scheme. Torsion-freeness over a valuation ring gives flatness by [StacksAlg] (Tag 0539). Each affine ring is of finite type over OO, as a quotient of the coordinate ring of the product. It is consequently finitely presented by [StacksAlg] (Tag 053E(1)). The closure is projective as a closed subscheme of the projective product; in particular it is quasi-compact and separated. The local finite presentation just proved is therefore a finite presentation.

The projections are proper, being restrictions of the proper product projections to a closed subscheme. On the generic fibre they are the given identities. If a target is integral, their closed images contain its dense generic fibre, proving surjectivity and birationality.

Base change to L∘L^\circ preserves projectivity, flatness and finite presentation, and gives generic fibre CLC_L. Theorem 4.13.2 identifies the actual completed generic fibre with CLanC_L^{\mathrm{an}}. Lemma 4.28.1 identifies its special fibre and the completed projection maps with their coefficient extensions. Surjectivity and functoriality of specialization are Proposition 3.17.2. No reducedness or nodality of the graph’s special fibre is needed for these assertions. ◻

Theorem 4.46.2. NAG-002311 (Algebraic domination from vertex inclusion). For two semistable algebraic models X1,X2\mathcal X_1,\mathcal X_2 of CC, there is an OO-morphism f:X1⟶X2f:\mathcal X_1\longrightarrow\mathcal X_2 inducing the identity on CC if and only if V(X2)⊂V(X1)V(\mathcal X_2)\subset V(\mathcal X_1) in CanC^{\mathrm{an}}. When it exists, it is unique and proper, and f∗OX1=OX2f_*\mathcal O_{\mathcal X_1}=\mathcal O_{\mathcal X_2}.

Proof. A morphism of the indicated models is proper by its closed graph and the properness of the source over OO. Its completion induces the analytic identity, so Lemma 4.40.4 proves the necessary vertex inclusion.

Suppose conversely that the inclusion holds. Form G\mathcal G and its two projections as in Lemma 4.46.1. Both targets are normal and integral by Proposition 4.44.3. We show that p2p_2 is constant on every set-theoretic fibre of p1p_1.

Choose the isometric comparison field Ω\Omega of Lemma 4.45.1, with residue field k‾\overline k. The two base-changed target models are semistable algebraic models over Ω∘\Omega^\circ. Their nodes have the split rational pointed charts of Theorem 4.18.2, since Ω\Omega is algebraically closed. Completing these charts gives the pointedly semistable formal models of Definition 4.39.4. By Corollary 4.39.5, the fibres of their specialization maps are their singleton component vertices and the connected components of the complements of those vertex sets. Proposition 4.45.2 identifies their vertex sets with the entire inverse images of the original vertex sets, so V(X2,Ω∘)⊂V(X1,Ω∘).V(\mathcal X_{2,\Omega^\circ}) \subset V(\mathcal X_{1,\Omega^\circ}).

Consequently sp⁡2\operatorname{sp}_2 is constant on every fibre of sp⁡1\operatorname{sp}_1 on CΩanC_\Omega^{\mathrm{an}}. For a component generic point the latter fibre is a singleton. For a closed point it is a connected component of the complement of the larger vertex set. It avoids the smaller vertex set, and lies in one connected component of its complement, which is one fibre of sp⁡2\operatorname{sp}_2.

Let z1,z2∈Gk‾z_1,z_2\in\mathcal G_{\overline k} have the same (p1)k‾(p_1)_{\overline k}-image. The surjective specialization in Lemma 4.46.1 supplies x1,x2∈CΩanx_1,x_2\in C_\Omega^{\mathrm{an}} with sp⁡G(xj)=zj\operatorname{sp}_{\mathcal G}(x_j)=z_j. Functoriality gives sp⁡1(x1)=sp⁡1(x2)\operatorname{sp}_1(x_1)=\operatorname{sp}_1(x_2). The preceding constancy gives sp⁡2(x1)=sp⁡2(x2)\operatorname{sp}_2(x_1)=\operatorname{sp}_2(x_2), hence (p2)k‾(z1)=(p2)k‾(z2)(p_2)_{\overline k}(z_1)=(p_2)_{\overline k}(z_2).

Here is the descent of this actual fibre relation. A field extension is flat by [StacksAlg] (Tag 00HD), since the only ideals of a field are zero and the whole field; its nonzero closed fibre makes it faithfully flat by [StacksAlg] (Tag 00HP). If t1,t2∈Gst_1,t_2\in\mathcal G_s have the same p1p_1-image yy, the tensor product κ(t1)⊗κ(y)κ(t2)\kappa(t_1)\otimes_{\kappa(y)}\kappa(t_2) is therefore nonzero. A prime of it, supplied by [StacksAlg] (Tag 00E0(2)), gives a point of Gs×X1,sGs\mathcal G_s\times_{\mathcal X_{1,s}}\mathcal G_s with projections t1,t2t_1,t_2. Faithful flatness of k‾/k\overline k/k and surjectivity on spectra, [StacksAlg] (Tag 00HQ), lift this point to the base-changed fibre product. Its two projections have the same (p1)k‾(p_1)_{\overline k}-image, and thus the same (p2)k‾(p_2)_{\overline k}-image. Projection back shows p2(t1)=p2(t2)p_2(t_1)=p_2(t_2). On the generic fibre both projections are identities. The generic and special fibres exhaust the rank-one base, and their images cannot meet. This proves the required constancy on every fibre of p1p_1.

Lemma 4.44.4, applied to the proper birational map p1p_1 and its normal target, gives a unique algebraic factorization p2=f∘p1p_2=f\circ p_1. It induces the identity on CC. Any two morphisms X1→X2\mathcal X_1\to \mathcal X_2 inducing that identity are equal: their equalizer is closed because X2/O\mathcal X_2/O is separated, and contains the schematically dense generic fibre of the flat integral source. Finally the same proper-quotient lemma applied to ff gives f∗OX1=OX2f_*\mathcal O_{\mathcal X_1}=\mathcal O_{\mathcal X_2}. ◻

Corollary 4.46.3. NAG-002312 (Algebraic uniqueness without a supplied common model). Two semistable algebraic models of CC with the same component vertices are uniquely isomorphic by an isomorphism inducing the identity on CC. The category of all semistable algebraic models of CC, with morphisms inducing the identity on CC, is equivalent to the partially ordered set of vertex sets that occur for such models, ordered by reverse inclusion.

If both models carry finite étale smooth markings identified with the same finite étale OO-scheme DD, with the same generic marking on CC, every morphism in this category respects these identifications.

Proof. The two vertex inclusions give morphisms in both directions by Theorem 4.46.2. Uniqueness in that theorem makes their composites identities. The same theorem identifies all morphism sets with the corresponding inclusion relation; composition agrees because the morphisms are unique.

For the marking assertion, compare the two maps from DD to the target. Their generic restrictions agree. On each affine open of the finite étale scheme DD, flatness makes its ring inject into its generic localization. The generic fibre of DD is therefore schematically dense. The target is separated, so the equalizer of the two maps is closed and must be all of DD. This proves compatibility with the specified marking isomorphisms. ◻

4.47 Unramified markings and common models

Definition 4.47.1. (Unramified marking data). An unramified marking datum on CC is a finite étale OO-scheme DD together with a closed immersion DK↪CD_K\hookrightarrow C. It is not required to extend to a smooth marking on any model chosen in advance.

Lemma 4.47.2. NAG-002315 (Descent of invariant ideals). Let L/KL/K be the finite unramified Galois lift of a finite Galois extension of kk, and put O′=L∘O'=L^\circ and G=Gal⁡(L/K)G=\operatorname{Gal}(L/K). Let X\mathcal X be an OO-scheme. Every GG-invariant quasi-coherent ideal J\mathcal J of finite type on XO′\mathcal X_{O'}, for the canonical coefficient action, is the extension of a unique quasi-coherent ideal I\mathcal I of finite type on X\mathcal X. Its blowup is the coefficient extension of the blowup of I\mathcal I. If J\mathcal J contains πn\pi^n, then I\mathcal I contains πn\pi^n.

Proof. Use the basis 1,α,…,αd−11,\alpha,\ldots,\alpha^{d-1} of O′/OO'/O from Lemma 4.43.1. The matrix (g(αi))g,i(g(\alpha^i))_{g,i} is invertible over O′O', by its unit Vandermonde determinant. On an affine open Spec⁡A⊂X\operatorname{Spec}A\subset\mathcal X, put A′=A⊗OO′A'=A\otimes_O O' and let J⊂A′J\subset A' be the given invariant ideal. If z=∑i=0d−1aiαi∈Jz=\sum_{i=0}^{d-1}a_i\alpha^i\in J, then g(z)=∑i=0d−1aig(αi)(g∈G).g(z)=\sum_{i=0}^{d-1}a_i g(\alpha^i) \qquad(g\in G). Inverting the matrix expresses each aia_i as an O′O'-linear combination of the g(z)g(z). Thus each aia_i belongs to I=J∩AI=J\cap A, and J=IA′J=IA'.

Choose finitely many generators of JJ. All their coefficients belong to II, and generate an ideal I0⊂AI_0\subset A with I0A′=JI_0A'=J. The coefficient-of-11 map A′→AA'\to A is AA-linear and is the identity on AA. Applying it to a finite expression for any a∈J∩Aa\in J\cap A proves a∈I0a\in I_0. Hence I=I0I=I_0 is finitely generated. The same coefficient map shows that extension of any ideal of AA contracts back to itself. This proves uniqueness.

On a principal open of Spec⁡A\operatorname{Spec}A, coefficient extraction commutes with localization. The ideals just constructed therefore agree on an affine basis of every overlap and glue to I\mathcal I. The assertion about πn\pi^n follows by contraction. Finally O′/OO'/O is faithfully flat by Lemma 4.43.1, and [StacksAlg] (Tag 0805) identifies the two blowups. This identification is induced by the coefficient map on their Rees algebras, and respects the canonical GG-actions. ◻

Theorem 4.47.3. NAG-002316 (Realizing an unramified marking on the original model). Let X\mathcal X be a semistable algebraic model of CC, and let DD be an unramified marking datum. A finite sequence of algebraic blowups supported in the special fibre gives a semistable algebraic model X′→X\mathcal X'\to\mathcal X in which DKD_K extends to a closed immersion of DD into the relative smooth locus. Every original component vertex is retained.

Proof. Choose a finite unramified Galois extension L/KL/K which splits the nodes of X\mathcal X and the finite étale scheme DsD_s. This can be done in the construction of Theorem 4.18.2: enlarge its chosen finite Galois residue extension to contain the splitting fields of the finitely many separable residue fields of DsD_s, and then take the lift of Lemma 4.18.1. The same pointed-chart construction works over this enlarged lift. Write O′=L∘O'=L^\circ and G=Gal⁡(L/K)G=\operatorname{Gal}(L/K). This lift satisfies [L:K]=[L~:k][L:K]=[\widetilde L:k] and ∣L×∣=∣K×∣|L^\times|=|K^\times|, with the given absolute value. The residue fields in this choice are finite separable by [StacksAlg] (Tag 02GL).

Every rational point of DL~D_{\widetilde L} extends to a section of DO′D_{O'} by Lemma 4.15.1. These sections are open and closed, since DO′D_{O'} is finite and étale. They are disjoint: two distinct sections cannot meet generically, since their equalizer is open and closed in the connected local spectrum and their closed points differ. Their complement is finite over O′O' and has empty closed fibre. Its finite coordinate algebra MM satisfies M/mO′M=0M/\mathfrak m_{O'}M=0, so it is zero by [StacksAlg] (Tag 00DV(2)). Thus DO′D_{O'} is a disjoint union of copies of Spec⁡O′\operatorname{Spec}O', and DLD_L gives a finite GG-invariant set of distinct LL-rational points of CLC_L.

Apply Theorem 4.34.3 to XO′\mathcal X_{O'} and this set. Its proof gives a finite sequence of blowups of GG-invariant finite-type ideals. Its Hausdorff generic-fibre hypothesis follows from Theorem 4.13.2 and Proposition 2.33.30. Each ideal is the finite product of the intrinsic ideals of the nodal or smooth marked clusters. Their support is a finite set of closed special points, and they contain a power of the original π\pi. At a nodal cluster, the actual charts are those of Lemma 4.27.1; at a smooth cluster they are those of Lemma 4.27.3. In particular these are actual flat nodal blowups over O′O', including when the value group is nondiscrete and when a node has both branches on the same component.

Descend the first ideal by Lemma 4.47.2 and blow it up over OO. Its coefficient extension is the first prescribed blowup, with the same GG-action. Repeat on the resulting original-field scheme. Induction descends the entire sequence. Every descended blowup is projective by [StacksAlg] (Tag 02NS), and its composite with the original projective model is projective by [StacksAlg] (Tag 0C4P). Its generic fibre is unchanged because its centre contains a power of π\pi. The nodal relative-curve property, including flatness and local finite presentation, descends from O′O' by [StacksAlg] (Tag 0C5E). Projectivity gives the required quasi-compactness and separatedness, hence finite presentation. The geometric special fibre remains connected: the displayed local charts replace a smooth point by an attached projective line, or a node by a chain of one or two projective lines. These replacements preserve connectedness after every residue-field extension. This proves that all descended stages are semistable algebraic models.

On the last coefficient extension, the marked sections give a GG-invariant closed immersion DO′↪XO′′D_{O'}\hookrightarrow\mathcal X'_{O'}. Its ideal is of finite type: locally at each marked point it is the ideal of a section in a smooth relative curve, as proved in Lemma 4.34.2; the sections are disjoint. Descend this ideal by Lemma 4.47.2, and write Z↪X′Z\hookrightarrow\mathcal X' for the resulting closed subscheme. The morphism Z/OZ/O is finite by [StacksAlg] (Tag 02LA), since its coefficient extension is DO′D_{O'}. The equivariant isomorphism ZO′≃DO′Z_{O'}\simeq D_{O'} identifies their finite coordinate algebras. Taking invariants and using Lemma 4.43.1 gives an OO-isomorphism Z≃DZ\simeq D, compatible with the generic marking. Its image is in the relative smooth locus. Indeed, the cover XO′′→X′\mathcal X'_{O'}\to\mathcal X' is étale. Around every marked point upstairs the structural map to Spec⁡O′\operatorname{Spec}O' is smooth, and Spec⁡O′→Spec⁡O\operatorname{Spec}O'\to\operatorname{Spec}O is étale. Smooth locality on the source, [StacksAlg] (Tag 036U), first gives smoothness of the composite and then descends it along this cover at the marked image. Finally Theorem 4.46.2 applied to X′→X\mathcal X'\to\mathcal X gives retention of all original component vertices. ◻

Lemma 4.47.4. NAG-002317 (Three marked branches force a vertex). Assume the ground field is algebraically closed. Let X\mathcal X be a semistable algebraic model of a smooth projective curve, and let PP be a finite set of rational points of its generic fibre whose sections have pairwise distinct smooth specializations. Suppose v∈Canv\in C^{\mathrm{an}} admits three pairwise disjoint connected open subsets B1,B2,B3B_1,B_2,B_3 with v∉⋃iBiv\notin\bigcup_i B_i such that Bi‾=Bi∪{v},Bi∩P≠∅.\overline{B_i}=B_i\cup\{v\}, \qquad B_i\cap P\ne\varnothing. Then vv is a component vertex of X\mathcal X.

Proof. Suppose vv is not a vertex, and let UU be its connected component in the complement of the model vertices. The split pointed charts of Theorem 4.18.2 complete to the pointed formal charts of Definition 4.39.4. Corollary 4.39.5 and Proposition 4.39.3 identify UU with an actual open-disc or open-annulus tube. Its boundary has one point in the disc case and at most two in the annulus case. Every BiB_i meets UU, since v∈U∩Bi‾v\in U\cap\overline{B_i} and UU is open.

If BiB_i is not contained in UU, it meets ∂U\partial U. Indeed, otherwise Bi∩UB_i\cap U would be a nonempty proper subset both open and closed in the connected space BiB_i. The disjoint BiB_i meet distinct boundary points when they meet ∂U\partial U. Thus at most one of them is not contained in a disc tube, and at most two are not contained in an annular tube. In the first case UU contains at least two members of PP, contradicting their distinct specializations. In the second it contains at least one member of PP, contradicting smoothness of its marked specialization. These contradictions prove the assertion. ◻

Theorem 4.47.5. NAG-002318 (Common original-field semistable models). Any finite nonempty family of semistable algebraic models of CC has a semistable algebraic model dominating all of them by morphisms inducing the identity on CC. If the models are marked by the same unramified marking datum DD, the common model and all the morphisms may be chosen to preserve that marking.

Proof. First consider two models X\mathcal X and Y\mathcal Y. Lemma 4.44.1 gives a finite étale smooth marking AA on Y\mathcal Y with at least three geometric points on every geometric special component, all distinct and away from the nodes. If a marking DD is already present, choose AA disjoint from it. Apply Theorem 4.47.3 to X\mathcal X and the generic marking datum AA, or A⨿DA\amalg D in the marked case. This gives a model Z→X\mathcal Z\to\mathcal X where these markings have pairwise distinct smooth specializations.

Compare over the isometric algebraically closed field Ω\Omega of Lemma 4.45.1. At a component vertex ww of YΩ∘\mathcal Y_{\Omega^\circ}, choose three of the auxiliary marked points on its geometric component. Their distinct smooth specializations give three disjoint open-disc tubes, each with closure equal to that disc together with ww, by Proposition 4.39.3. They contain the chosen generic marked points. Lemma 4.47.4, applied to ZΩ∘\mathcal Z_{\Omega^\circ}, makes ww a component vertex there. Proposition 4.45.2 then gives the original-field inclusion V(Y)⊂V(Z)V(\mathcal Y)\subset V(\mathcal Z). Theorem 4.46.2 produces the actual algebraic morphism Z→Y\mathcal Z\to\mathcal Y. In the marked case both morphisms respect DD by Corollary 4.46.3.

Induct on the finite number of models, replacing the first two by their common model. Each new common model dominates the preceding one, so composition retains every morphism already constructed, as well as the common marking. ◻

4.48 An intrinsic arithmetic condition on vertices

Fix the isometric algebraically closed comparison field Ω\Omega of Lemma 4.45.1, and write p:CΩan→Canp:C_\Omega^{\mathrm{an}}\to C^{\mathrm{an}} for the projection.

Definition 4.48.1. NAG-002320 (Unramified branch certificates). A point v∈Canv\in C^{\mathrm{an}} has an unramified branch certificate if there are a point w∈p−1(v)w\in p^{-1}(v), an unramified marking datum PP, and three pairwise disjoint open analytic discs B1,B2,B3⊂CΩanB_1,B_2,B_3\subset C_\Omega^{\mathrm{an}}, with w∉⋃iBiw\notin\bigcup_iB_i, such that Bi‾=Bi∪{w},Bi∩PK(Ω)≠∅(1≤i≤3).\overline{B_i}=B_i\cup\{w\}, \qquad B_i\cap P_K(\Omega)\ne\varnothing \quad(1\leq i\leq3). Closures are taken in CΩanC_\Omega^{\mathrm{an}}.

Theorem 4.48.2. NAG-002321 (Which original-field points occur as model vertices). Suppose CC admits a semistable algebraic model over OO. A point of CanC^{\mathrm{an}} occurs as a component vertex of such a model if and only if it has an unramified branch certificate. Every finite set of points having such certificates is contained in the component vertex set of one original-field semistable algebraic model dominating any initially specified model. The certificate condition is independent of the chosen algebraically closed comparison field.

For every such point vv, its entire inverse image under pp is finite and consists of type-II points. Moreover ∣H(v)×∣=∣K×∣,H(v)~/k is separably generated of transcendence degree one.\begin{aligned} &|\mathcal H(v)^\times|=|K^\times|,\\ &\widetilde{\mathcal H(v)}/k \text{ is separably generated of transcendence degree one}. \end{aligned}

Proof. Suppose first that vv is the vertex belonging to a special component EE of a model Y\mathcal Y. Lemma 4.44.1 supplies an unramified marking PP on Y\mathcal Y, meeting every geometric component over EE at least three times in distinct smooth points. Choose one geometric component over EE and its vertex ww. Proposition 4.45.2 gives p(w)=vp(w)=v. The tubes of three of those marked specializations are disjoint open discs whose closures adjoin precisely ww, by Proposition 4.39.3. They contain the corresponding points of PK(Ω)P_K(\Omega) and give the certificate.

Conversely, start with any specified semistable algebraic model X\mathcal X. For a certificate (w,P,B1,B2,B3)(w,P,B_1,B_2,B_3), Theorem 4.47.3 gives a model X′→X\mathcal X'\to\mathcal X on which PP is a disjoint smooth marking. After coefficient extension to Ω\Omega, Lemma 4.47.4 makes ww a component vertex of XΩ∘′\mathcal X'_{\Omega^\circ}. By Proposition 4.45.2, its image vv is an original component vertex of X′\mathcal X'. For finitely many certificates, repeat this construction one at a time. Each refinement retains all preceding vertices, so the final model contains every prescribed point and dominates the initial model.

Apply Proposition 4.45.2 to any model just obtained. It proves the displayed arithmetic equalities and identifies the entire inverse image with its finitely many geometric component vertices. These are type-II points by Corollary 4.39.5. The construction of a certificate from a model works for every comparison field, proving the asserted independence. ◻

Definition 4.48.3. (Admissible original-field vertex sets). Fix an unramified marking datum DD, possibly empty. A finite nonempty subset V⊂CanV\subset C^{\mathrm{an}} is admissible for (C,D)(C,D) if every point of VV has an unramified branch certificate and V‾=p−1(V)\overline V=p^{-1}(V) is a finite semistable vertex set of CΩanC_\Omega^{\mathrm{an}} with the following marking condition: every open-disc component of CΩan∖V‾C_\Omega^{\mathrm{an}}\setminus\overline V contains at most one point of DK(Ω)D_K(\Omega), and every open-annulus component contains none.

Theorem 4.48.4. NAG-002323 (Intrinsic correspondence over the original field). Suppose CC admits at least one semistable algebraic model over OO, and fix an unramified marking datum DD. The component vertex map induces a bijection between isomorphism classes of semistable algebraic models of CC carrying DD as a finite étale smooth marking and the finite nonempty vertex sets admissible for (C,D)(C,D) in Definition 4.48.3. For two such models there is a morphism inducing the identity on CC and respecting DD precisely when the target vertex set is contained in the source vertex set; that morphism is unique. Thus this bijection is an equivalence with the partially ordered set under reverse inclusion. Admissibility is independent of the chosen comparison field.

Proof. Let Y\mathcal Y be such a marked model. Its vertices have certificates by Theorem 4.48.2. Proposition 4.45.2 identifies their entire inverse image with its geometric component vertices. After extension to Ω\Omega, the pointed charts and Corollary 4.39.5 make this a finite semistable vertex set. Distinct marked sections have distinct smooth specializations, so its open-disc tubes contain at most one marked point and its nodal annular tubes contain none. This proves admissibility.

Conversely, let VV be admissible. By Theorem 4.48.2, choose an original-field semistable algebraic model whose vertex set contains VV. Refine it by Theorem 4.47.3 to make DD a disjoint smooth marking. Write (X,D)(\mathcal X,D) for the resulting marked model and WW for its vertex set. It still has V⊂WV\subset W. Over Ω\Omega, the entire inverse images satisfy V‾⊂W‾\overline V\subset\overline W by Proposition 4.45.2.

Consider a connected component TT of the geometric dual subgraph on the vertices outside V‾\overline V. Since V‾\overline V is semistable, Lemma 4.41.2 identifies TT with the part of the source skeleton in one component UU of CΩan∖V‾C_\Omega^{\mathrm{an}}\setminus\overline V. It is a finite tree, and each of its geometric component curves itself is a smooth Pk‾1\mathbf P^1_{\overline k}, hence has no self-node. The number of its attaching branches is one if UU is an open disc and two if UU is an open annulus, with both branches counted even when they attach to the same protected component.

Every marking whose specialization is on a component of TT belongs to UU. Indeed, its smooth X\mathcal X tube is connected, avoids V‾\overline V, and has that component vertex in its closure. Since this vertex lies in open UU, the tube meets UU and therefore lies in that connected component of the complement. The marking condition for VV consequently bounds the number of marked legs on TT by one in the disc case and by zero in the annular case. Thus TT satisfies precisely the geometric marked forest condition of Definition 4.45.4.

Theorem 4.45.5 now gives an actual original-field marked model with component vertices exactly VV. Its generic identification is the original one and its marking is isomorphic to the prescribed DD. Corollary 4.46.3 proves uniqueness of the marked model, and Theorem 4.46.2, together with the marking assertion of that corollary, identifies all its morphisms. This proves the correspondence and its categorical assertion. Its characterization by original-field models also proves independence of the comparison field. ◻

Corollary 4.48.5. NAG-002324 (Extension of semilinear actions). Let a group GG act on CC and an unramified marking datum DD by compatible semilinear isometric automorphisms. If VV is an invariant admissible vertex set and CC admits a semistable algebraic model over OO, the model corresponding to VV carries a unique GG-action inducing the prescribed generic action and preserving the marking. All morphisms arising from inclusions between invariant admissible vertex sets are equivariant.

Proof. An isometric field automorphism preserves OO. For g∈Gg\in G, pull the model back by its base automorphism and use its prescribed generic curve isomorphism to identify the new generic fibre with CC. This is again a semistable algebraic model. Functorial specialization identifies its vertices with the corresponding image of VV, which is VV. The pulled-back marking is identified with DD by its given semilinear action. Corollary 4.46.3 therefore gives the unique semilinear model isomorphism extending gg. Two composites which have the same generic action are equal by the schematic-density equalizer argument in the proof of Theorem 4.46.2. Thus these isomorphisms satisfy the group law and form the asserted action. The same uniqueness, applied to the two composites of a vertex-inclusion morphism with gg, proves equivariance of every such morphism. ◻

4.49 Least semistable refinements over the original field

Keep the original field KK, valuation ring OO, curve CC, and model category of Section 4.46. Fix the comparison field Ω\Omega of Lemma 4.45.1, and put k‾=Ω~\overline k=\widetilde\Omega. All vertex labels below are the actual analytic component points. In particular, retaining a label retains its point and its original absolute value.

Lemma 4.49.1. NAG-002326 (Geometric components over one original component). Let YY be a scheme of finite type over a field kk, and let k‾\overline k be an algebraic closure. The geometric irreducible components over any one irreducible component of YY form one orbit under Aut⁡(k‾/k)\operatorname{Aut}(\overline k/k). Consequently an invariant subset of the geometric components is the entire inverse image of a subset of the original components.

Proof. Let ksep⊂k‾k^{\mathrm{sep}}\subset\overline k be the separable closure of kk in k‾\overline k. The assertion over ksepk^{\mathrm{sep}} is [StacksAlg] (Tag 04KY). In characteristic zero this is already the assertion. In characteristic p>0p>0, the remaining extension k‾/ksep\overline k/k^{\mathrm{sep}} is purely inseparable: an irreducible polynomial over kk has the form Q(Tpe)Q(T^{p^e}) with QQ separable, by successively removing zero derivatives. A root therefore has a pep^e-th power separable over kk.

For a ksepk^{\mathrm{sep}}-algebra AA, write B=A⊗ksepk‾B=A\otimes_{k^{\mathrm{sep}}}\overline k. This is a faithfully flat extension: a field extension is flat by the finite-ideal criterion [StacksAlg] (Tag 00HD), is faithfully flat by [StacksAlg] (Tag 00HP), and its tensor functor on AA-modules is the same coefficient tensor functor. Thus every prime of AA has a prime above it by [StacksAlg] (Tag 00HQ). For every b∈Bb\in B, some pp-power bpeb^{p^e} belongs to AA: write bb as a finite coefficient sum and choose ee simultaneously for its finitely many coefficients. For a prime QQ above PP, one therefore has b∈Q⟺bpe∈P.b\in Q\quad\Longleftrightarrow\quad b^{p^e}\in P. This determines QQ uniquely and preserves inclusions in both directions. Hence minimal primes, and consequently irreducible components, correspond. These correspondences commute with affine restriction and with automorphisms of ksep/kk^{\mathrm{sep}}/k. Every such automorphism extends uniquely to k‾\overline k: the image of aa is the unique pep^e-th root of the prescribed image of ape∈ksepa^{p^e}\in k^{\mathrm{sep}}. The component orbit assertion over ksepk^{\mathrm{sep}} therefore gives the assertion over k‾\overline k. The final statement follows by taking unions of these orbits. ◻

Definition 4.49.2. (Protected marked graphs and elementary pruning). A protected marked graph is a finite connected multigraph GG, with loops allowed, a finite set of labelled legs at its vertices, a subset RR of its vertices, and a nonempty subset SS of protected vertices. A loop contributes two to the degree deg⁡G(v)\deg_G(v); write m(v)m(v) for the number of marked legs at vv. A vertex is deletable if v∉S,v∈R,v has no loop,deg⁡G(v)+m(v)≤2.v\notin S,\qquad v\in R,\qquad v\text{ has no loop},\qquad \deg_G(v)+m(v)\leq2. Its degree is positive by connectedness and S≠∅S\ne\varnothing. For a degree-one deletable vertex, remove its incident edge and transfer its zero or one marked leg to the neighbouring vertex. For a degree-two deletable vertex, remove it and replace its two edges by one edge between the other ends; this may create a loop. Such a vertex has no marked leg. Retain the vertex labels, the membership in RR, and the protected labels at every other vertex. These operations are called elementary pruning.

Lemma 4.49.3. NAG-002328 (Confluence and finite forests). Every sequence of elementary pruning terminates. Any sequence continued until no deletable vertex remains has the same labelled terminal graph, up to an incidence isomorphism fixing its retained vertices and its labelled legs.

Suppose AA is a subset of the unprotected vertices such that each connected component TT of the induced subgraph on AA is a tree, all its vertices belong to RR, and b(T)+m(T)≤2,b(T)+m(T)\leq2, where b(T)b(T) counts edges from TT to its complement and m(T)m(T) counts its marked legs. Contracting each such tree, transferring its possible mark in the one-ended case and joining its two ends in the two-ended case, is a sequence of elementary prunings. In particular it does not change the terminal graph.

Proof. Every pruning removes one vertex and retains the nonempty set SS, so a sequence is finite. We verify a local commutation property. If two distinct deletable vertices are not adjacent, their operations commute. Their incident edges are disjoint; even when they have a common neighbour, the transferred labelled legs and the replacement edges give the same result in either order. Each vertex remains deletable after the other operation.

Suppose instead that the two vertices are adjacent. They have only one edge between them. Two parallel edges would exhaust both degrees, making them the whole connected graph, contrary to the existence of a protected vertex. For the same reason they cannot both have degree one. If one is a leaf and the other has degree two, deleting the leaf leaves a leaf with at most one marked leg at the second vertex. Deleting the second vertex first joins the original leaf directly to the remaining neighbour. Both orders remove the pair and transfer the same possible leg to that neighbour. If both vertices have degree two, neither has a mark, and both orders suppress the same two-vertex path. Their other neighbours may coincide: both orders then give the same loop at that common neighbour. In all cases both second deletions remain permissible and the results agree.

Induct on the number of vertices to prove confluence. For two terminal sequences with the same first deletion, apply induction to the smaller graph. For different first deletions, the preceding commutation property gives a common graph after two deletions. Induction applied to each of the two one-deletion graphs identifies its terminal graph with the terminal graph of this common graph. Thus the two original terminal graphs agree. This also proves that their retained labels and transferred marked-leg labels agree.

For the forest assertion, each TT has b(T)≥1b(T)\geq1, because the whole graph is connected and its complement contains SS. A one-vertex tree is immediately deletable. If TT has more than one vertex, it has at least two leaves: take the two ends of a longest simple path. Since its total number of external ends and marks is at most two, some leaf carries at most one of these external ends or marks. Its one internal edge therefore gives deg⁡G(v)+m(v)≤2\deg_G(v)+m(v)\leq2. Delete it. If it carries an external edge, that edge is joined to its internal neighbour; if it carries a mark, that mark is transferred to that neighbour. The remaining tree has the same external ends and marked legs, and its vertices still belong to RR. Continue until the tree disappears. Its final operation is exactly the prescribed one-ended or two-ended contraction, including a loop when the two external ends have the same neighbour. Different connected components of AA have no edge between them, so this procedure applies to all of them. Confluence proves the last assertion. ◻

Proposition 4.49.4. NAG-002329 (The least contraction retaining specified vertices). Let (X,D)(\mathcal X,D) be a marked semistable algebraic model, with vertex set WW, and let ∅≠S⊂W\varnothing\ne S\subset W. There is a marked semistable algebraic model XS\mathcal X_S and a morphism X→XS\mathcal X\to\mathcal X_S retaining SS such that every marked contraction X→Y\mathcal X\to\mathcal Y retaining SS has a unique factorization X⟶Y⟶XS.\mathcal X\longrightarrow\mathcal Y \longrightarrow\mathcal X_S. Its geometric marked graph is the terminal graph of Definition 4.49.2, with protected set equal to the entire inverse image of SS and with RR the vertices whose normalized residual component is Pk‾1\mathbf P^1_{\overline k}. The construction uses finitely many actual contractions over OO.

Proof. Use the actual coefficient extension to Ω\Omega from Lemma 4.28.1 and the geometric marked dual graph of Definition 4.45.3. Its vertex set is the entire inverse image of WW by Proposition 4.45.2. The normalized residual component at a retained geometric vertex is unchanged under a model morphism: it is the intrinsic normal proper curve of Theorem 4.29.2. This makes the decoration RR independent of subsequent contractions.

A geometric component is itself a smooth projective line exactly when its normalization is a projective line and it has no self-node. Here a self-node is precisely a loop of the dual graph. Indeed the actual normalization squares of Lemma 4.24.1 separate exactly the two branches of each node. At a node joining different components, the selected component has just one smooth branch: in a pointed étale chart it is one of the two axes. Smooth locality on the source, [StacksAlg] (Tag 036U), descends this assertion to that component. Away from the nodes normalization is already an isomorphism, as in Lemma 4.31.1. Thus a component without self-nodes is its smooth normalization; the assertion follows.

Let AA be the set of all currently deletable geometric vertices. It is invariant under Aut⁡(k‾/k)\operatorname{Aut}(\overline k/k): the action preserves incidence, loops, the marked-leg set, the normalized component isomorphism class, and the protected set. Lemma 4.49.1 makes AA the entire inverse image of a subset of original components.

Every vertex of the graph induced by AA has degree at most two in the full graph. A connected component TT of it cannot contain a cycle: all vertices of that cycle would already have full degree two and no edge leaving it. It would be the whole connected graph, contrary to the nonempty protected complement. Thus TT is a tree. Put a=∣T∣a=|T|. Counting its internal edges gives b(T)+m(T)=∑v∈T(deg⁡G(v)+m(v))−2(a−1)≤2.b(T)+m(T) =\sum_{v\in T}\bigl(\deg_G(v)+m(v)\bigr)-2(a-1) \leq2. All its actual geometric components are smooth projective lines by the preceding paragraph. Therefore the complement of AA satisfies the exact marked forest condition, and Theorem 4.45.5 constructs an actual original-OO contraction deleting precisely AA. Its fibre description gives exactly the graph operations in Lemma 4.49.3, including the transfer of a mark and the creation of a self-node from two coincident ends. The whole marking remains the original finite étale OO-scheme.

Repeat with all deletable vertices of the new model. Each nonempty round removes original components and retains SS, so after finitely many rounds the process stops at a model XS\mathcal X_S with no deletable geometric vertex. Each round is a forest contraction and hence decomposes combinatorially into elementary prunings by Lemma 4.49.3. That lemma identifies the resulting graph with its unique terminal graph. Single geometric deletions in this combinatorial comparison need not descend individually; every algebraic round used the entire invariant forest and the original-field contraction theorem.

Now let q:X→Yq:\mathcal X\to\mathcal Y be any marked contraction retaining SS. Proposition 4.45.6 says that its removed components satisfy the marked forest condition. The target has the graph of that forest contraction: the exact contraction of Theorem 4.45.5 to the same vertex set has the stated fibres and is identified with qq by Proposition 4.45.7. It is consequently a sequence of elementary prunings on the geometric graph. Continuing to a terminal graph and applying confluence shows that every retained vertex of XS\mathcal X_S is a vertex of Y\mathcal Y. The entire-fibre identification of Proposition 4.45.2 gives the same inclusion for original analytic vertices. Theorem 4.46.2 now supplies the actual map Y→XS\mathcal Y\to\mathcal X_S, and Corollary 4.46.3 makes it marked. Uniqueness of the morphisms inducing the generic identity proves the factorization and its uniqueness. ◻

Theorem 4.49.5. NAG-002330 (The least marked semistable hull). Suppose CC admits a semistable algebraic model over OO, and fix an unramified marking datum DD. For every finite nonempty set S⊂CanS\subset C^{\mathrm{an}} of points with unramified branch certificates, there is a least finite vertex set HD(S)H_D(S) admissible for (C,D)(C,D) and containing SS. Equivalently, there is a marked semistable algebraic model MD(S)\mathcal M_D(S) containing SS such that every marked semistable algebraic model containing SS has a unique marked morphism to MD(S)\mathcal M_D(S) inducing the identity on CC.

For S⊂S′S\subset S' the resulting models have a unique compatible morphism MD(S′)→MD(S)\mathcal M_D(S')\to\mathcal M_D(S). Any finite nonempty family of models marked by DD has a least common dominating model: take SS to be the union of their vertex sets. Any group of compatible semilinear isometric automorphisms of (C,D)(C,D) preserving SS extends uniquely to MD(S)\mathcal M_D(S).

Proof. Theorem 4.48.2 constructs an original-field semistable model containing SS. Apply Theorem 4.47.3 to give it the prescribed smooth marking DD, while retaining SS; call the resulting model X\mathcal X. Proposition 4.49.4 gives M=XS\mathcal M=\mathcal X_S.

Let Y\mathcal Y be any other model marked by DD and containing SS. Theorem 4.47.5 gives an actual common marked semistable model T\mathcal T of X\mathcal X and Y\mathcal Y. Its geometric graph can be pruned to that of X\mathcal X, and then to the terminal graph of M\mathcal M, by Proposition 4.45.6 and Lemma 4.49.3. It can also be pruned to the graph of Y\mathcal Y. The protected geometric labels are the same entire inverse image of SS throughout, by Proposition 4.45.2. Confluence therefore makes the terminal graph obtained from Y\mathcal Y equal to that of M\mathcal M, with the same retained analytic labels. In particular V(M)⊂V(Y)V(\mathcal M)\subset V(\mathcal Y). Theorem 4.46.2 and Corollary 4.46.3 give the unique marked map Y→M\mathcal Y\to\mathcal M.

The vertex set HD(S)=V(M)H_D(S)=V(\mathcal M) is admissible by Theorem 4.48.4. Conversely every admissible finite set containing SS is the vertex set of a model in the preceding comparison. Thus it contains HD(S)H_D(S). This proves both the minimum and independence of the initially chosen model. It also proves that M\mathcal M is unique up to its unique generic-identity marked isomorphism.

If S⊂S′S\subset S', the model for S′S' contains SS, giving the stated unique morphism and all composition compatibilities. For a finite family of marked models, each vertex has a certificate by Theorem 4.48.2. Apply the construction to their finite nonempty union. Its model dominates every family member by vertex inclusion. Every common dominating model contains that union and therefore has a unique map to this one.

Finally a compatible semilinear isometry preserving SS transports the class of marked models containing SS to itself. The universal property identifies its transform of M\mathcal M with M\mathcal M; equivalently, it preserves HD(S)H_D(S). Corollary 4.48.5 extends these isometries to the model and proves the group law and uniqueness. Every construction kept the original valuation ring and absolute value; Ω\Omega was used only for the geometric graph comparison. ◻

4.50 Integral Gauss charts and arithmetic branch certificates

Let KK be complete and nontrivially valued for its specified real-valued absolute value. Put O=K∘O=K^\circ, k=K~k=\widetilde K, and choose 0≠π∈K∘∘0\ne\pi\in K^{\circ\circ}. For a polynomial g∈O[T]g\in O[T] with nonzero reduction, put R0=O⟨T,Z⟩/(g(T)Z−1),R=R0[1/π].R_0=O\langle T,Z\rangle/(g(T)Z-1),\qquad R=R_0[1/\pi]. Proposition 3.17.1 identifies R0R_0 with the principal formal open of the integral unit disc. Its special fibre is Rˉ=k[t,1/gˉ]\bar R=k[t,1/\bar g], and its generic fibre is the actual domain ∣T∣≤1|T|\leq1, ∣g(T)∣=1|g(T)|=1.

Definition 4.50.1. (Integral étale Gauss charts). An integral étale Gauss chart is a nonzero admissible OO-algebra A0A_0 which is finite projective as an R0R_0-module and is an étale R0R_0-algebra. Write A=A0[1/π]A=A_0[1/\pi], U=M(A)U=\mathcal M(A) and Y=Spec⁡(A0/K∘∘A0)Y=\operatorname{Spec}(A_0/K^{\circ\circ}A_0). The following propositions use these notations for a fixed chart.

Proposition 4.50.2. NAG-002333 (The actual Gauss fibres of an integral chart). The special fibre YY is a reduced pure curve. For each irreducible component EE of YY, its function field k(E)k(E) is finite separable over k(t)k(t). Let η\eta be the unit Gauss point of M(R)\mathcal M(R) and let H=H(η)H=\mathcal H(\eta) with its specified absolute value. The actual fibre of U→M(R)U\to\mathcal M(R) over η\eta consists exactly of the component points vEv_E. For every such point, [H(vE):H]=[k(E):k(t)],H(vE)~=k(E),∣H(vE)×∣=∣K×∣.\begin{aligned} [\mathcal H(v_E):H]&=[k(E):k(t)],\\ \widetilde{\mathcal H(v_E)}&=k(E),\qquad |\mathcal H(v_E)^\times|=|K^\times|. \end{aligned} More precisely, its actual integral fibre algebra satisfies A0⊗R0H∘≃∏EHE∘,A_0\otimes_{R_0}H^\circ \simeq\prod_E H_E^\circ, where HE/HH_E/H is a finite unramified lift of k(E)/k(t)k(E)/k(t). After inverting π\pi, this is the decomposition of the actual analytic fibre algebra into its completed residue fields.

Proof. Put B=A0/K∘∘A0B=A_0/K^{\circ\circ}A_0. It is finite projective and étale over the domain Rˉ\bar R; étaleness commutes with base change by [StacksAlg] (Tag 02GO). As a direct summand of a finite free module, BB injects into its localization B⊗Rˉk(t)=∏i=1rli,B\otimes_{\bar R}k(t)=\prod_{i=1}^r l_i, where the li/k(t)l_i/k(t) are finite separable fields by [StacksAlg] (Tag 02GL). Thus BB is reduced. Let PiP_i be the kernel of its map to lil_i. These prime ideals have intersection zero and are pairwise incomparable: localization gives the distinct factor primes in the displayed product. Every prime of BB contains one of the PiP_i, because the product of these finitely many ideals is zero. They are consequently exactly the minimal primes. Moreover Frac⁡(B/Pi)=li\operatorname{Frac}(B/P_i)=l_i: localization by the nonzero elements of Rˉ\bar R already gives lil_i. By [StacksAlg] (Tag 00P0), every component has dimension one. This proves the special-fibre assertions and identifies the index ii with its component EE.

The weighted Gauss calculation of Proposition 1.2.2 gives ∣g(η)∣=1|g(\eta)|=1. The field HH is the completion of K(T)K(T) for its unit Gauss norm. Indeed, polynomials and their fractions are dense in the completed residue field; adjoining 1/g1/g does not change that dense rational subfield. A quotient of two polynomials has norm in ∣K×∣|K^\times|. If it has norm one, dividing numerator and denominator by coefficients of their respective maximal sizes identifies its residue with a rational function of tt. Conversely every rational function of tt arises this way. Completion changes neither values nor residues by Lemma 3.22.2. Hence H~=k(t),∣H×∣=∣K×∣.\widetilde H=k(t),\qquad |H^\times|=|K^\times|. Evaluation gives an OO-homomorphism R0→H∘R_0\to H^\circ.

Set D0=A0⊗R0H∘D_0=A_0\otimes_{R_0}H^\circ. This algebra is finite projective and étale over H∘H^\circ. By [StacksAlg] (Tag 00NX), its module is free, because H∘H^\circ is local. Its reduction is D0/H∘∘D0=∏Ek(E).D_0/H^{\circ\circ}D_0=\prod_E k(E). For each factor, Lemma 4.15.2 provides a complete finite extension HE/HH_E/H with residue field k(E)k(E), unchanged value group, and degree [k(E):k(t)][k(E):k(t)]. Lemma 4.22.1 makes HE∘H_E^\circ finite free over H∘H^\circ with that rank. The projection of the displayed product to k(E)k(E) is a rational point of the special fibre of Spec⁡(D0⊗H∘HE∘)\operatorname{Spec}(D_0\otimes_{H^\circ}H_E^\circ). The henselian section lemma 4.15.1 gives a homomorphism D0→HE∘D_0\to H_E^\circ reducing to that projection. Combining these homomorphisms gives D0⟶∏EHE∘.D_0\longrightarrow\prod_E H_E^\circ. Both sides are finite free of rank ∑E[k(E):k(t)]\sum_E[k(E):k(t)]: on the left this rank is read by reduction of a basis. Its matrix has invertible reduction, so its determinant is a unit of the local ring H∘H^\circ. The adjugate matrix proves that it is an isomorphism.

Choose an R0R_0-linear splitting A0⟶R0m⟶A0A_0\longrightarrow R_0^m\longrightarrow A_0. The integral presentations of Proposition 3.14.7 identify A0A_0 and R0R_0 with the closed unit balls of their presentation norms. Both split maps preserve these unit balls. Scaling by powers of π\pi makes both maps bounded after inversion, so the given norm on AA is equivalent to its canonical finite-RR-module quotient norm. Proposition 1.18.7 identifies A⊗^RH=A⊗RH=D0[1/π]=∏EHE.A\widehat\otimes_RH=A\otimes_RH=D_0[1/\pi] =\prod_E H_E. Theorem 2.33.11 identifies this with the actual fibre over η\eta, including its completed residue fields. Each factor has just its specified valued point, by uniqueness of the absolute value in Lemma 1.12.1. Its specialization in YY is the component generic point of EE, by the reduction of the map just constructed. Proposition 4.12.2 identifies that point with vEv_E, and gives uniqueness above each component generic point. The degree, residue-field and value-group equalities follow from the construction of HEH_E. ◻

Lemma 4.50.3. NAG-002334 (Unramified closed points from separable residual points). Suppose UU is identified with an actual affinoid domain of the analytification of a smooth projective geometrically connected curve C/KC/K. Every closed point q∈Yq\in Y with l=κ(q)l=\kappa(q) finite separable over kk lifts to an OO-homomorphism A0⟶L∘,A_0\longrightarrow L^\circ, where L/KL/K is a finite unramified lift of l/kl/k. Its generic analytic point is the analytification of a closed point of CC with residue field exactly LL. Distinct residual points give distinct closed points of CC. The disjoint union of the corresponding Spec⁡L∘\operatorname{Spec}L^\circ for any finite such list is an unramified marking datum of Definition 4.47.1.

Proof. Choose L/KL/K by Lemma 4.15.2. Its valuation ring has the monic presentation O[α]=O[S]/(P)O[\alpha]=O[S]/(P) and the orthonormal basis 1,α,…,α[l:k]−11,\alpha,\ldots,\alpha^{[l:k]-1} of Lemma 4.22.1. Since P′(α)P'(\alpha) is a unit, across a square-zero ideal the unique correction of a lifted root bb is b−P(b)/P′(b)b-P(b)/P'(b). This proves formal étaleness; the finite presentation is étale by [StacksAlg] (Tag 00UR). Thus L∘/OL^\circ/O is finite free and étale, with the original extending norm.

Choose c∈L∘c\in L^\circ lifting the image of tt at qq. The value g(c)g(c) is a unit because gˉ(q)≠0\bar g(q)\ne0. Substitution T↦cT\mapsto c, Z↦g(c)−1Z\mapsto g(c)^{-1} gives R0→L∘R_0\to L^\circ. The finite étale algebra A0⊗R0L∘A_0\otimes_{R_0}L^\circ has the rational special point defined by qq. Lemma 4.15.1 provides a section through it, and hence the stated map. It is continuous for the π\pi-adic topology. Its generic map A→LA\to L is bounded by the integral-presentation norm of Proposition 3.14.7.

Let x∈U⊂Canx\in U\subset C^{\mathrm{an}} be the resulting point. On an affine algebraic neighbourhood of its image, the mapping property of Theorem 2.33.23 gives a homomorphism from its coordinate ring to LL. Its image is a finite-dimensional domain over KK, hence a field: the minimal polynomial of any nonzero element expresses its inverse as a polynomial in that element. Thus the corresponding point p∈Cp\in C is closed and κ(p)⊆L\kappa(p)\subseteq L. Proposition 2.33.28 and completeness of finite extensions in Lemma 1.12.1 identify H(x)\mathcal H(x) with κ(p)\kappa(p).

Choose a∈A0a\in A_0 whose residue at qq is a primitive generator β\beta of l/kl/k, using [StacksAlg] (Tag 030N). Write bb for its value in L∘L^\circ and e=[l:k]e=[l:k]. The change-of-basis matrix from 1,α,…,αe−11,\alpha,\ldots,\alpha^{e-1} to 1,b,…,be−11,b,\ldots,b^{e-1} has invertible reduction, since both residue lists are bases of l/kl/k. Its determinant is a unit, so 1,b,…,be−11,b,\ldots,b^{e-1} is an OO-basis of L∘L^\circ. Evaluation of analytic functions takes values in H(x)\mathcal H(x), with the same values under its canonical embedding in LL, by Proposition 1.6.2. Consequently b∈κ(p)b\in\kappa(p) and K(b)=LK(b)=L, proving κ(p)=L\kappa(p)=L.

Two different points qq have different analytic lifts, because specialization is a function. A finite residue field over the complete field KK has only its unique extending absolute value, so one closed algebraic point cannot give those two different analytic points. The resulting maps Spec⁡L→C\operatorname{Spec}L\to C are therefore distinct closed immersions. Their finite disjoint union is a closed immersion, while their integral coefficient schemes are finite étale over OO. This is exactly the required marking datum. ◻

Theorem 4.50.4. NAG-002335 (Integral Gauss charts give branch certificates). Suppose UU is an actual affinoid domain of CanC^{\mathrm{an}} as in Lemma 4.50.3. Every component point vEv_E of YY has an unramified branch certificate. If CC admits a semistable algebraic model over the original OO, then every vEv_E occurs as a component vertex of such a model. Finitely many points obtained from finitely many such integral charts can be retained simultaneously on one original-field semistable algebraic model.

Proof. Fix the algebraically closed comparison field Ω\Omega of Lemma 4.45.1. Put RΩ,0=Ω∘⟨T,Z⟩/(g(T)Z−1),AΩ,0=A0⊗R0RΩ,0.R_{\Omega,0}=\Omega^\circ\langle T,Z\rangle/(g(T)Z-1), \qquad A_{\Omega,0}=A_0\otimes_{R_0}R_{\Omega,0}. The first algebra is admissible by Proposition 3.17.1. The second is finite projective over it, hence complete: an idempotent makes it a closed direct summand of a finite free complete module. It is π\pi-torsion-free and topologically of finite type, and is therefore admissible by Theorem 3.14.4. For the finite-type assertion, finite module generators give finitely many algebra generators; reducing monomials by their integral equations bounds all their products in a finite module, so restricted series converge. Étaleness is preserved by [StacksAlg] (Tag 02GO).

The actual scalar presentation 2.33.5 identifies RΩ,0[1/π]=R⊗^KΩR_{\Omega,0}[1/\pi]=R\widehat\otimes_K\Omega. Finite tensor comparison 1.18.7 consequently identifies AΩ,0[1/π]=A⊗^KΩA_{\Omega,0}[1/\pi]=A\widehat\otimes_K\Omega. The domain base-change theorem 2.36.9 makes its spectrum the actual domain UΩ⊂CΩanU_\Omega\subset C_\Omega^{\mathrm{an}}. Its special fibre is YΩ~Y_{\widetilde\Omega}. Proposition 4.50.2, applied over Ω\Omega, makes it a reduced pure curve.

Choose a component E′E' of YΩ~Y_{\widetilde\Omega} above EE. Such a component exists, and the component fibres are exactly those described in Lemma 4.49.1. Let ww be its component point. Functoriality of specialization in Proposition 3.17.2 and uniqueness in Proposition 4.12.2 show that ww maps to vEv_E.

Choose three closed separable points qiq_i of YY and geometric lifts qi′q_i' lying on E′E', with the qiq_i pairwise distinct. Here is a construction over the original residue field. Inside Ω~\widetilde\Omega take the separable closure ksepk^{\mathrm{sep}} of kk. The purely inseparable step in Lemma 4.49.1 identifies the components over these two fields. In the component corresponding to E′E', take a nonempty open subset avoiding the other components and any finitely many previously selected Galois orbits. Its étale map to the principal open of Aksep1\mathbb A^1_{k^{\mathrm{sep}}} has open image by [StacksAlg] (Tag 03WT). The field ksepk^{\mathrm{sep}} is infinite: a finite field of cardinality qq has the separable polynomial Xq−X−1X^q-X-1 without a root. Thus that image contains a rational point. Its nonempty étale fibre has a rational point by [StacksAlg] (Tag 02GL). It is defined over a finite separable extension of kk, because the algebra has finitely many generators. Its image in YY is a closed separable point. Repeating this three times gives the required distinct qiq_i and qi′q_i'.

Use Lemma 4.50.3 to form the unramified marking datum P=∐iSpec⁡Li∘P=\coprod_i\operatorname{Spec}L_i^\circ. Each prescribed residue embedding L~i→Ω~\widetilde L_i\to\widetilde\Omega lifts to an isometric KK-embedding Li→ΩL_i\to\Omega. Indeed, in the monic presentation of Lemma 4.22.1, the prescribed residual root is simple. The henselian root construction 4.15.1 in Ω∘\Omega^\circ lifts it to a root of the same polynomial. Irreducibility over KK gives the embedding, and uniqueness in Lemma 1.12.1 makes it isometric. The resulting member of PK(Ω)P_K(\Omega) specializes to qi′q_i'.

Let Bi=]qi′[B_i=]q_i'[ in Spf⁡AΩ,0\operatorname{Spf}A_{\Omega,0}. The formal map to the principal open of the unit disc is adic, topologically finitely presented and formally étale. Composition with the formal open immersion into the disc has the same properties: unique lifting composes, and the relative presentation is obtained by adjoining ZZ with g(T)Z=1g(T)Z=1 to a finite presentation. Completion gives the restricted-series presentation by Lemma 3.14.8. Translate the disc coordinate by a lift of t(qi′)t(q_i'). Proposition 4.39.3 now makes BiB_i an actual open unit disc whose closure in UΩU_\Omega is Bi∪{w}B_i\cup\{w\}. In particular the selected marked point lies in BiB_i, and the three tubes are disjoint.

These discs are open also in CΩanC_\Omega^{\mathrm{an}}. To verify this ordinary-topological assertion, fix x∈Bix\in B_i and an ambient affinoid neighbourhood WW of xx. The disc is an analytic domain and is absolutely boundaryless: in its disc coordinate, every point belongs to the strict interior of some smaller closed disc. Since WW is an ambient ordinary neighbourhood, Bi∩WB_i\cap W is an ordinary neighbourhood of xx in BiB_i. Choose an affinoid neighbourhood VV of xx in BiB_i with V⊂Bi∩WV\subset B_i\cap W. Proposition 2.44.6 gives x∈Int⁡(V/Ω)x\in\operatorname{Int}(V/\Omega). The composition formula 2.44.1 for V→W→M(Ω)V\to W\to\mathcal M(\Omega) then gives x∈Int⁡(V/W)x\in\operatorname{Int}(V/W). The domain-interior theorem 2.44.3 makes VV a neighbourhood of xx in WW, and hence in CΩanC_\Omega^{\mathrm{an}}. This proves openness. Finally UΩU_\Omega is compact and closed in the Hausdorff curve CΩanC_\Omega^{\mathrm{an}}, by Propositions 2.24.1 and 2.33.29. Thus the same closure formula holds in the ambient curve. The data w,P,B1,B2,B3w,P,B_1,B_2,B_3 form the certificate of Definition 4.48.1.

If an original-field semistable model exists, the two remaining assertions are precisely the realization and finite simultaneous-insertion assertions of Theorem 4.48.2. ◻

4.51 Integral Gauss charts at component vertices

Keep the complete nontrivially real-valued field KK, its specified absolute value, O=K∘O=K^\circ, k=K~k=\widetilde K, and 0≠π∈K∘∘0\ne\pi\in K^{\circ\circ}. All completions are π\pi-adic.

Lemma 4.51.1. NAG-002337 (Lifting residual finiteness). Let R0→A0R_0\to A_0 be a continuous homomorphism of admissible OO-algebras. Suppose that A0A_0 is topologically of finite type over R0R_0 and that A0/K∘∘A0is finite overR0/K∘∘R0.A_0/K^{\circ\circ}A_0 \quad\hbox{is finite over}\quad R_0/K^{\circ\circ}R_0. Then A0A_0 is a finite R0R_0-module.

Proof. Choose a continuous surjection P=R0⟨Y1,…,Ys⟩↠A0P=R_0\langle Y_1,\ldots,Y_s\rangle\twoheadrightarrow A_0, and write yiy_i for the image of YiY_i. Each yˉi\bar y_i satisfies a monic polynomial over Rˉ=R0/K∘∘R0\bar R=R_0/K^{\circ\circ}R_0. Indeed, write multiplication by yˉi\bar y_i on a finite list of module generators as a matrix over Rˉ\bar R. Multiplication by the adjugate of XI−MX I-M shows that its monic determinant annihilates every generator, hence also 11. Choose such a polynomial of positive degree did_i, and lift its coefficients to a monic Pi(X)∈R0[X]P_i(X)\in R_0[X]. Then Pi(yi)∈K∘∘A0P_i(y_i)\in K^{\circ\circ}A_0.

Each of these finitely many memberships is a finite sum of elements of A0A_0 multiplied by scalars in K∘∘K^{\circ\circ}. Choose one nonzero θ∈K∘∘\theta\in K^{\circ\circ} whose absolute value is at least those of all these scalars; if they all vanish, take θ=π\theta=\pi. Divisibility in OO gives Pi(yi)=θqiP_i(y_i)=\theta q_i with qi∈A0q_i\in A_0. Lift qiq_i to Qi∈PQ_i\in P and put Fi=Pi(Yi)−θQi.F_i=P_i(Y_i)-\theta Q_i. These elements map to zero in A0A_0.

Give R=R0[1/π]R=R_0[1/\pi] the complete presentation norm whose closed unit ball is R0R_0, as in Proposition 3.14.7. On polynomials over RR, division by the monic Pi(Yi)P_i(Y_i) does not increase the coefficient norm: its lower coefficients belong to R0R_0. Perform these divisions in the order Y1,…,YsY_1,\ldots,Y_s. Their quotient and remainder operators are RR-linear of norm at most one. Polynomial density and completeness extend them to R⟨Y1,…,Ys⟩R\langle Y_1,\ldots,Y_s\rangle, and they preserve PP. Thus there are continuous R0R_0-linear operators DiD_i and rr such that h=∑iPi(Yi)Di(h)+r(h),r(h)∈L:=∑0≤αi<diR0Yα.h=\sum_i P_i(Y_i)D_i(h)+r(h),\qquad r(h)\in L:=\sum_{0\leq\alpha_i<d_i}R_0Y^\alpha. The displayed finite direct sum LL is complete. Set E(h)=θ∑iQiDi(h)E(h)=\theta\sum_i Q_iD_i(h). Its norm is at most ∣θ∣ ∥h∥|\theta|\,\|h\|, so h=∑iFiDi(h)+r(h)+E(h).h=\sum_i F_iD_i(h)+r(h)+E(h). Iteration and convergence give h=∑iFi(∑j≥0Di(Ej(h)))+∑j≥0r(Ej(h)).h=\sum_i F_i\left(\sum_{j\geq0}D_i(E^j(h))\right) +\sum_{j\geq0}r(E^j(h)). Both sums converge in the indicated integral modules: their terms have norms at most ∣θ∣j∥h∥|\theta|^j\|h\|, and powers of θ\theta eventually lie in every prescribed power of πO\pi O. The last sum belongs to LL. Consequently the finitely many images of YαY^\alpha, 0≤αi<di0\leq\alpha_i<d_i, generate A0A_0 over R0R_0. ◻

Lemma 4.51.2. NAG-002338 (Finite étale integral lifting). Let R0→A0R_0\to A_0 be as in Lemma 4.51.1. Suppose in addition that A0/πA0A_0/\pi A_0 is an étale R0/πR0R_0/\pi R_0-algebra. Then A0A_0 is finite projective as an R0R_0-module and is an étale R0R_0-algebra.

Proof. The preceding lemma makes A0A_0 a finite R0R_0-module. It has no OO-torsion, so Theorem 3.14.4 makes this module finitely presented. Its reduction modulo π\pi is therefore a finitely presented module. It is flat by [StacksAlg] (Tag 02GS), and hence finite projective by [StacksAlg] (Tag 00NX). Choose an idempotent matrix eˉ∈Mm(R0/πR0)\bar e\in M_m(R_0/\pi R_0) whose image is this module.

Lift eˉ\bar e to a matrix e0e_0 over R0R_0. If δ=e2−e∈πMm(R0)\delta=e^2-e\in\pi M_m(R_0), then (2e−I)2=I+4δ(2e-I)^2=I+4\delta is invertible by its convergent geometric series. The Newton correction e′=e−(2e−I)−1(e2−e)e'=e-(2e-I)^{-1}(e^2-e) has error e′2−e′e'^2-e' divisible by δ2\delta^2. All matrices in this calculation commute, being convergent expressions in ee. Starting at e0e_0 gives a convergent sequence with an idempotent limit e≡eˉ(modπ)e\equiv\bar e\pmod\pi. This works in every characteristic. Put P=eR0mP=eR_0^m. It is a complete separated direct summand of R0mR_0^m.

Lift the images of the standard basis under the map (R0/πR0)m→A0/πA0(R_0/\pi R_0)^m\to A_0/\pi A_0 to A0A_0, and restrict the resulting map R0m→A0R_0^m\to A_0 to PP. The map φ:P→A0\varphi:P\to A_0 is an isomorphism modulo π\pi. For a∈A0a\in A_0, lift its class modulo π\pi to p0∈Pp_0\in P, write a−φ(p0)=πa1a-\varphi(p_0)=\pi a_1, and repeat. Completeness shows that ∑n≥0πnpn\sum_{n\geq0}\pi^np_n maps to aa. Thus φ\varphi is surjective. If φ(p)=0\varphi(p)=0, the modulo-π\pi isomorphism gives p=πp1p=\pi p_1 in PP. Since A0A_0 has no π\pi-torsion, φ(p1)=0\varphi(p_1)=0. Iteration puts pp in ⋂nπnP=0\bigcap_n\pi^nP=0. Thus A0≃PA_0\simeq P is finite projective over R0R_0.

The algebra A0A_0 is finitely presented over R0R_0. For completeness, choose a finite module presentation with generators b1,…,bmb_1,\ldots,b_m. Express 11 and all products bibjb_ib_j in these generators. The finite module relations, the relation for 11, and these product relations present A0A_0 as an algebra: they reduce every polynomial to a linear combination of the bib_i, whose kernel is exactly the module relation space. It follows from the same polynomial presentation that ΩA0/R0\Omega_{A_0/R_0} is a finite R0R_0-module and that its reduction modulo π\pi is Ω(A0/πA0)/(R0/πR0)=0\Omega_{(A_0/\pi A_0)/(R_0/\pi R_0)}=0. For every a∈R0a\in R_0, 1−πa1-\pi a is invertible by a geometric series. Thus π\pi belongs to the Jacobson radical of R0R_0, and [StacksAlg] (Tag 00DV) gives ΩA0/R0=0\Omega_{A_0/R_0}=0. Finite projectivity implies flatness by [StacksAlg] (Tag 00NX); the flat, finitely presented, zero-differential criterion [StacksAlg] (Tag 02GU) now proves ordinary étaleness. ◻

Theorem 4.51.3. NAG-002339 (Integral Gauss-chart recognition of model vertices). Let C/KC/K be smooth, projective and geometrically connected. Every component vertex of a semistable algebraic model of CC over the original OO, in the sense of Definition 4.45.3, is the component point of an integral étale Gauss chart whose generic fibre is an actual affinoid domain of CanC^{\mathrm{an}}. The special fibre of this chart may be chosen integral.

Consequently, if CC admits one such original-field model, the following conditions on v∈Canv\in C^{\mathrm{an}} are equivalent:

  1. vv is a component vertex of a semistable algebraic model over OO;

  2. vv is a component point of an integral étale Gauss chart, as in Definition 4.50.1, whose generic fibre is an actual affinoid domain of CanC^{\mathrm{an}};

  3. vv has an unramified branch certificate of Definition 4.48.1.

Proof. Let C\mathcal C be the given model and let ηE\eta_E be the generic point of its special component EE. The special nodal curve is smooth at every component generic point. Flatness, finite presentation, and [StacksAlg] (Tag 01V9) therefore make the family smooth at ηE\eta_E. By [StacksAlg] (Tag 054L) there is an affine open neighbourhood V=Spec⁡B\mathcal V=\operatorname{Spec}B of ηE\eta_E with an étale coordinate map V→AO1\mathcal V\to\mathbf A^1_O. The smooth family has relative dimension one, so the affine-space dimension in that lemma is one. Shrink V\mathcal V by a principal open so that its special fibre is integral. To do so, in the reduced finite-type kk-algebra B/K∘∘BB/K^{\circ\circ}B, there are only finitely many minimal primes by [StacksAlg] (Tags 00FO and 00FR). Choose, for every minimal prime other than that of EE, an element in that prime but not in the prime of EE. Their product vanishes on all other components and is nonzero at ηE\eta_E. Lift it to BB and invert it. The remaining special fibre is a nonempty open of EE. Its reduced coordinate ring has a unique minimal prime, which is zero by [StacksAlg] (Tag 00E0), so it is a domain. The coordinate map remains étale.

Write Bˉ=B/K∘∘B\bar B=B/K^{\circ\circ}B after this shrink, and write tt for the residual coordinate. The map Spec⁡Bˉ→Ak1\operatorname{Spec}\bar B\to\mathbf A^1_k is étale by [StacksAlg] (Tag 02GO); its image is nonempty open by [StacksAlg] (Tag 03WT). Hence k[t]→Bˉk[t]\to\bar B is injective. Its generic fibre is a nonzero domain étale of finite type over k(t)k(t), so [StacksAlg] (Tag 02GL) makes it a finite separable field. That field is k(E)k(E).

Choose finitely many k[t]k[t]-algebra generators of Bˉ\bar B. Each satisfies a monic polynomial over k(t)k(t). Clearing the finitely many coefficient denominators gives one nonzero gˉ∈k[t]\bar g\in k[t] for which Bˉ[1/gˉ]is finite overk[t,1/gˉ].\bar B[1/\bar g] \quad\hbox{is finite over}\quad k[t,1/\bar g]. Indeed, each chosen generator is now integral, and the monic relations reduce their powers to finitely many bounded-exponent monomials. Choose a lift g∈O[T]g\in O[T]. The algebra B[1/g(T)]B[1/g(T)] remains finitely presented and flat over OO, and is étale of finite presentation over O[T,1/g(T)]O[T,1/g(T)].

Let A0A_0 be its actual π\pi-adic completion, and set R0=O⟨T,Z⟩/(g(T)Z−1).R_0=O\langle T,Z\rangle/(g(T)Z-1). Lemma 4.13.1 gives admissibility, the actual restricted presentation, and unchanged reductions modulo every πn\pi^n; the base completion is R0R_0 by Proposition 3.17.1. The relative algebraic presentation gives a continuous surjection R0⟨Y1,…,Ys⟩→A0R_0\langle Y_1,\ldots,Y_s\rangle\to A_0. Because π∈K∘∘\pi\in K^{\circ\circ}, its reduction modulo K∘∘K^{\circ\circ} is exactly Bˉ[1/gˉ]\bar B[1/\bar g]. Thus the residual-finiteness hypothesis of Lemma 4.51.1 holds. Its reduction modulo π\pi is the base change of the ordinary étale algebraic coordinate map, and is étale by [StacksAlg] (Tag 02GO). Lemma 4.51.2 proves that A0A_0 is finite projective and étale over R0R_0. Its nonempty integral special fibre is the chosen open of EE.

The completion of C\mathcal C is an actual formal model of CanC^{\mathrm{an}} by Theorem 4.13.2. The completed affine open just constructed has generic fibre U=M(A0[1/π])U=\mathcal M(A_0[1/\pi]), an actual affinoid domain of CanC^{\mathrm{an}}; completion and restriction agree by Lemma 4.13.1 and Proposition 3.17.2. Its special generic point is still ηE\eta_E. The component-point uniqueness and restriction statement of Proposition 4.12.2 therefore identify its unique component point with the original vertex vEv_E. This proves the first assertion.

It gives (1)⇒\Rightarrow(2). Theorem 4.50.4 gives (2)⇒\Rightarrow(3), and Theorem 4.48.2 gives (3)⇒\Rightarrow(1) under the stated existence hypothesis. All constructions use OO, its chosen π\pi, and its specified absolute value throughout. ◻

4.52 A numerical vertex criterion over discretely valued fields

Except in statements explicitly requiring discreteness, KK is any complete nontrivially real-valued field, with its specified absolute value, O=K∘O=K^\circ and k=K~k=\widetilde K. Choose 0≠π∈K∘∘0\ne\pi\in K^{\circ\circ} for formal completions. Write η\eta for the unit Gauss point of the analytic affine line and H=H(η)H=\mathcal H(\eta). Thus HH is the Gauss completion of K(T)K(T), with residue field k(t)k(t) and value group ∣K×∣|K^\times|, as computed in Proposition 4.50.2.

Lemma 4.52.1. NAG-002341 (An integral basis when the discrete value group is unchanged). Let HH be a complete discretely valued field and let L/HL/H be a finite extension with its extending absolute value. If ∣L×∣=∣H×∣|L^\times|=|H^\times|, then [L:H]=[L~:H~].[L:H]=[\widetilde L:\widetilde H]. Every lift b1,…,bf∈L∘b_1,\ldots,b_f\in L^\circ of a H~\widetilde H-basis of L~\widetilde L is an orthonormal HH-basis of LL and an H∘H^\circ-basis of L∘L^\circ. No separability assumption on the residue extension is needed.

Proof. Any finite residue-independent list has linearly independent lifts over HH. In a nonzero relation, divide its coefficients by one having maximal absolute value and reduce. At least one coefficient has nonzero residue, contrary to residue independence. Consequently f=[L~:H~]f=[\widetilde L:\widetilde H] is finite and at most [L:H][L:H]: otherwise one could choose an independent residual list longer than that finite dimension. The same argument gives ∣∑j=1fcjbj∣=max⁡j∣cj∣(cj∈H).\left|\sum_{j=1}^f c_jb_j\right| =\max_j|c_j|\qquad(c_j\in H).

Let ϖ\varpi be a uniformizer of HH. The equality of value groups implies L∘∘=ϖL∘L^{\circ\circ}=\varpi L^\circ. For a∈L∘a\in L^\circ, express its residue in the given basis and lift the coefficients to H∘H^\circ. Subtracting the corresponding linear combination gives ϖa1\varpi a_1 with a1∈L∘a_1\in L^\circ. Repeat on a1,a2,…a_1,a_2,\ldots. For each basis vector the resulting coefficient is a convergent series ∑n≥0ϖncn,j\sum_{n\geq0}\varpi^n c_{n,j} in H∘H^\circ. The residual error has norm at most ∣ϖ∣n+1|\varpi|^{n+1}, so its limit is zero. This expresses aa in ∑jH∘bj\sum_jH^\circ b_j. Any element of LL becomes integral after multiplication by a power of ϖ\varpi. Thus these ff vectors span LL, and the displayed norm identity proves both the degree equality and the asserted integral basis. ◻

Lemma 4.52.2. NAG-002342 (Principal formal domains inside a Gauss neighbourhood). Let g∈O[T]g\in O[T] have nonzero reduction, and put Dg={x:∣T(x)∣≤1, ∣g(x)∣=1}.D_g=\{x:|T(x)|\leq1,\ |g(x)|=1\}. Every ordinary open neighbourhood NN of η\eta in DgD_g contains DghD_{gh} for some h∈O[T]h\in O[T] with nonzero reduction.

Proof. By Proposition 3.17.1, DgD_g is the actual principal formal generic domain, with its subspace topology in the closed unit disc. The evaluation topology of Proposition 1.6.3 therefore supplies finitely many functions fi∈K⟨T⟩f_i\in K\langle T\rangle and an ordinary open neighbourhood N′N' of η\eta in that disc, contained in the chosen relative neighbourhood after intersection with DgD_g, whose membership is ensured by keeping the values ∣fi∣|f_i| in specified open intervals containing ∣fi(η)∣|f_i(\eta)|. Zero functions impose no restriction and can be omitted.

For nonzero fif_i, its Gauss norm is attained by a coefficient ai∈K×a_i\in K^\times. By the Gauss calculation and polynomial density in Proposition 1.2.2, choose hi∈O[T]h_i\in O[T] with ∥ai−1fi−hi∥<1,∥hi∥=1.\|a_i^{-1}f_i-h_i\|<1, \qquad\|h_i\|=1. In particular hˉi≠0\bar h_i\ne0. Put h=∏ihih=\prod_i h_i, with the empty product equal to one. On DghD_{gh} every factor hih_i has absolute value one: all have value at most one on the unit disc, and their product has value one. The strict approximation then gives ∣fi(x)∣=∣ai∣=∣fi(η)∣|f_i(x)|=|a_i|=|f_i(\eta)| throughout DghD_{gh}. Thus Dgh⊂N′∩Dg⊂ND_{gh}\subset N'\cap D_g\subset N. The reduction of ghgh is nonzero because k[t]k[t] is a domain. ◻

Proposition 4.52.3. NAG-002343 (An unramified Gauss germ gives an actual integral chart). Let XX be a good KK-analytic space, let v∈Xv\in X, and suppose that an analytic morphism from an ordinary neighbourhood of vv to the analytic affine line is étale at vv and sends vv to η\eta. Put L=H(v)L=\mathcal H(v), with the induced map H→LH\to L. Suppose that L/HL/H is finite, that l=L~l=\widetilde L is finite separable over k(t)k(t), and that [L:H]=[l:k(t)].[L:H]=[l:k(t)]. Then vv belongs to an actual affinoid domain of XX carrying an integral étale Gauss chart. Its special fibre is integral and vv is its unique component point.

Proof. Choose a primitive element α\alpha of l/k(t)l/k(t) by [StacksAlg] (Tag 030N), and let Pˉ∈k(t)[S]\bar P\in k(t)[S] be its monic minimal polynomial of degree ff. Choose g0∈O[T]g_0\in O[T] with nonzero reduction so that all coefficients of Pˉ\bar P lie in k[t,1/gˉ0]k[t,1/\bar g_0]. Lift them to O[T,1/g0]O[T,1/g_0], obtaining a monic polynomial PP. On the free module O[T,1/g0][S]/(P)O[T,1/g_0][S]/(P) with its monic remainder basis, let Δ∈O[T,1/g0]\Delta\in O[T,1/g_0] be the determinant of multiplication by P′P'. Its reduction is nonzero in k(t)k(t), because Pˉ\bar P is separable. Write Δ=a(T)/g0(T)m\Delta=a(T)/g_0(T)^m with a∈O[T]a\in O[T]; then aˉ≠0\bar a\ne0. Put g=g0ag=g_0a. The adjugate matrix makes P′P' a unit in B0alg=O[T,1/g][S]/(P).B^{\mathrm{alg}}_0=O[T,1/g][S]/(P). For an arbitrary square-zero ideal, a lifted root is corrected uniquely by b↦b−P(b)/P′(b)b\mapsto b-P(b)/P'(b). Thus this finite free algebra is formally étale, and is ordinarily étale by [StacksAlg] (Tag 00UR).

Let R0=O⟨T,Z⟩/(g(T)Z−1),B0=R0[S]/(P).R_0=O\langle T,Z\rangle/(g(T)Z-1),\qquad B_0=R_0[S]/(P). The algebra B0algB^{\mathrm{alg}}_0 is flat over OO: coefficientwise injectivity in the ideal criterion [StacksAlg] (Tag 00HD) makes O[T]O[T] flat over OO; localization is flat by [StacksAlg] (Tag 00HT), and the final extension is finite free. The actual completion calculation in Lemma 4.13.1 identifies B0B_0 with the completion of B0algB^{\mathrm{alg}}_0, coefficient by coefficient in its finite free monic basis. It is admissible, finite free and étale over R0R_0. Its special algebra is finite free over k[t,1/gˉ]k[t,1/\bar g] and injects into its generic fibre ll, so it is a domain. Consequently Proposition 4.50.2 gives exactly one point ww over η\eta in B=M(B0[1/π])B=\mathcal M(B_0[1/\pi]), with [H(w):H]=f[\mathcal H(w):H]=f and residual extension l/k(t)l/k(t). Here 0≠π∈K∘∘0\ne\pi\in K^{\circ\circ} is any choice used for the completion.

The residue element α∈L~\alpha\in\widetilde L is a simple root of Pˉ\bar P. The complete-valuation henselian lemma 4.15.1 lifts it to a root of PP in L∘L^\circ. It gives an HH-homomorphism H(w)=H[S]/(P)→L\mathcal H(w)=H[S]/(P)\to L. Both fields have degree ff over HH, so it is an isomorphism; it is isometric by Lemma 1.12.1.

Use the full algebraic generic cover Spec⁡B0alg[1/π]⟶Spec⁡K[T,1/g].\operatorname{Spec}B^{\mathrm{alg}}_0[1/\pi] \longrightarrow\operatorname{Spec}K[T,1/g]. Its analytification B\mathcal B is étale over the ordinary open g≠0g\ne0 in the analytic line by Proposition 2.33.24 and Theorem 2.35.2. Theorem 2.33.26 identifies BB with its entire inverse image of the affinoid domain DgD_g. In particular its entire fibre at η\eta is {w}\{w\}.

By the ordinary-open local definition of an analytic étale map, the map near vv restricts to a finite flat map with zero relative differentials. Finite branch separation in Lemma 2.35.1 then gives a finite étale germ with fibre field LL. The completed-residue-field equivalence of Theorem 2.54.3 therefore gives an isomorphism over the analytic line between ordinary neighbourhoods of vv in XX and of ww in B\mathcal B. Write WW for the latter neighbourhood.

By Proposition 1.6.3, the compact set B∖WB\setminus W has closed image in the Hausdorff affinoid DgD_g. Its image avoids η\eta because the full fibre is the singleton w∈Ww\in W. Lemma 4.52.2 supplies a polynomial h∈O[T]h\in O[T] with nonzero reduction such that the domain DghD_{gh} avoids that image. Its full inverse image in BB is therefore contained in WW. It is the actual affinoid generic fibre of the principal formal restriction B0⟨U⟩/(h(T)U−1)B_0\langle U\rangle/(h(T)U-1) by Proposition 3.17.1. This restriction is finite projective and étale over the principal base chart for ghgh, by ordinary base change and [StacksAlg] (Tag 02GO). Its special algebra is the corresponding nonzero localization of the integral special algebra of B0B_0. The ordinary-neighbourhood isomorphism transports its actual affinoid domain into XX. Its unique component point is transported from ww to vv, proving the assertion. ◻

Theorem 4.52.4. NAG-002344 (Discrete original-field vertex recognition). Suppose KK is complete and discretely valued, with its specified nontrivial absolute value. Let C/KC/K be a smooth projective geometrically connected curve. Let v∈Canv\in C^{\mathrm{an}} satisfy ∣H(v)×∣=∣K×∣,H(v)~/k is separably generated of transcendence degree one.\begin{aligned} &|\mathcal H(v)^\times|=|K^\times|,\\ &\widetilde{\mathcal H(v)}/k \text{ is separably generated of transcendence degree one}. \end{aligned} Then vv is the unique component point of an actual integral étale Gauss chart in CanC^{\mathrm{an}}, and has an unramified branch certificate. If CC admits one semistable algebraic model over the original OO, these two displayed conditions are necessary and sufficient for vv to occur as a component vertex of such a model.

Proof. Put L=H(v)L=\mathcal H(v) and l=L~l=\widetilde L. The scheme image of vv is the generic point of CC. Indeed, a nongeneric point on an integral curve is closed; its residue field is finite over KK by [StacksAlg] (Tag 00FV). Its analytic completed residue field is that same finite complete field by Proposition 2.33.28 and Lemma 1.12.1, whose residual field is algebraic over kk by Lemma 3.22.2. This contradicts the assumed residual transcendence degree. Thus F=K(C)F=K(C) embeds densely in LL, by the same analytification-valued-fibre proposition.

Choose tˉ∈l\bar t\in l as a separating transcendence basis, so that l/k(tˉ)l/k(\bar t) is algebraic separable. Density supplies t0∈Ft_0\in F of norm one with residue tˉ\bar t. Choose a rational smooth coordinate s∈Fs\in F from a smooth affine chart at the generic point, using [StacksAlg] (Tag 054L). In its standard smooth Jacobian presentation, solving the invertible square subsystem expresses every differential in terms of dsds; consequently ΩF/K\Omega_{F/K} has basis dsds. For a uniformizer ϖ\varpi of KK, choose a sufficiently large integer NN and set t=t0+ϖNs.t=t_0+\varpi^Ns. Then ∣ϖNs(v)∣<1|\varpi^Ns(v)|<1, so the residue of tt is still tˉ\bar t. We can require dt≠0dt\ne0: in the one-dimensional space ΩF/K\Omega_{F/K}, at most one scalar among the distinct ϖN\varpi^N makes dt0+ϖNdsdt_0+\varpi^Nds vanish.

The rational function tt is transcendental over KK. The finitely generated field F/KF/K has transcendence degree one, so it is a finite extension of K(t)K(t): its finite field generators are algebraic over K(t)K(t). The quotient of ΩF/K\Omega_{F/K} obtained by imposing dt=0dt=0 is ΩF/K(t)\Omega_{F/K(t)}, as follows directly from the universal property of derivations. It is zero. On an affine open where tt is regular, the map to the affine line is of finite presentation; this follows from [StacksAlg] (Tag 00FO) and a finite polynomial presentation. At its generic point the local map is the field extension K(t)→FK(t)\to F, which is flat by [StacksAlg] (Tag 00HD) because the only ideals of the base field are zero and itself. The criterion [StacksAlg] (Tag 02GU) therefore makes this map étale at the generic point. This algebraic open analytifies to an ordinary open by Proposition 2.33.24. Theorem 2.35.2 gives an actual analytic étale map near vv.

For a polynomial Q∈K[T]Q\in K[T], division by a coefficient of maximal size and reduction at vv give ∣Q(t)∣=max⁡i∣Qi∣,|Q(t)|=\max_i|Q_i|, since tˉ\bar t is transcendental over kk. Thus vv maps to the unit Gauss point, and completion embeds the Gauss field HH isometrically into LL. The image of F⊗K(t)H→LF\otimes_{K(t)}H\to L is a finite-dimensional HH-subalgebra of the field LL, containing FF. It is a field: multiplication by a nonzero element is injective on this finite-dimensional space and hence surjective. It is complete and therefore closed by Lemma 1.12.1. Since FF is dense in LL, this image is all of LL. In particular L/HL/H is finite.

The Gauss-field computation in Proposition 4.50.2 makes HH discretely valued with value group ∣K×∣|K^\times| and residue field k(tˉ)k(\bar t). The assumed value-group equality and Lemma 4.52.1 give [L:H]=[l:k(tˉ)].[L:H]=[l:k(\bar t)]. In particular the right side is finite; by the choice of tˉ\bar t, this residual extension is separable. Proposition 4.52.3 constructs the actual integral chart in CanC^{\mathrm{an}}. Theorem 4.50.4 supplies its unramified branch certificate.

When one original-field semistable model exists, realization follows from Theorem 4.48.2. Conversely, every component vertex of such a model has exactly the stated value group and separably generated residual function field by Lemma 4.45.2. The field, its absolute value and the uniformizer used in the proof have not been replaced by an extension or rescaling. ◻

4.53 An original-field rational Gauss degree criterion

Let KK be complete and nontrivially valued for its specified real-valued absolute value, and put O=K∘O=K^\circ and k=K~k=\widetilde K. Let C/KC/K be a smooth projective geometrically connected curve and put F=K(C)F=K(C). Semistable algebraic models are those of Definition 4.45.3.

Theorem 4.53.1. NAG-002346 (Rational Gauss degree criterion). Assume that CC admits one semistable algebraic model over the original OO. A point v∈Canv\in C^{\mathrm{an}} occurs as a component vertex of such a model if and only if there is a rational function t∈Ft\in F, defined at vv, satisfying the following conditions. Write L=H(v)L=\mathcal H(v) and l=L~l=\widetilde L.

  1. ∣t(v)∣=1|t(v)|=1, and its residue tˉ∈l\bar t\in l is transcendental over kk.

  2. F/K(t)F/K(t) and l/k(tˉ)l/k(\bar t) are finite separable field extensions.

  3. If HtH_t is the completion of K(t)K(t) for the norm induced by vv, then the actual extension Ht→LH_t\to L is finite and [L:Ht]=[l:k(tˉ)].[L:H_t]=[l:k(\bar t)].

Here HtH_t is the unit Gauss completion for the coordinate tt. The criterion applies without a discreteness or characteristic-zero assumption on KK.

Proof. Suppose first that vv is the component vertex of a model C\mathcal C, corresponding to the generic point ηE\eta_E of its special component EE. Use the construction in Theorem 4.51.3, retaining its original algebraic coordinate. More explicitly, that construction chooses an actual affine Zariski open of C\mathcal C at ηE\eta_E with an étale map to AO1\mathbf A^1_O. It removes the other special components by a principal open, makes the residual map finite by a principal shrink of the affine-line base, and completes that same algebraic open. The coordinate TT therefore restricts to a rational function t∈Ft\in F; it is not a newly chosen analytic coordinate. The resulting completed algebra is finite projective and étale over O⟨T,Z⟩/(g(T)Z−1),gˉ≠0,O\langle T,Z\rangle/(g(T)Z-1), \qquad \bar g\ne0, its special fibre is an open of EE, and its actual component point in CanC^{\mathrm{an}} is the original vv.

By Proposition 4.50.2, the point vv lies above the unit Gauss point in coordinate tt, its residual function field l=k(E)l=k(E) is finite separable over k(tˉ)k(\bar t), and its actual completed field degree is precisely this residual degree. In particular ∣t(v)∣=1|t(v)|=1 and tˉ\bar t is transcendental. The retained algebraic coordinate is also étale on the generic fibre by [StacksAlg] (Tag 02GO). Localizing its integral generic coordinate ring by all nonzero polynomials in tt gives a nonzero domain which is étale of finite type over K(t)K(t). By [StacksAlg] (Tag 02GL) it is a finite separable field, necessarily FF. This proves all three conditions with the actual original-field rational function tt.

Conversely, suppose the conditions hold. The scheme image of vv is generic: otherwise it is a closed point of the integral curve, its residue field is finite over KK by [StacksAlg] (Tag 00FV), and Proposition 2.33.28 and Lemma 1.12.1 identify LL with that finite complete field. Its residual field would be algebraic over kk by Lemma 3.22.2, contrary to the transcendence of tˉ\bar t. Thus all of FF embeds into LL with its specified norm.

On an affine algebraic open where tt is regular, the rational map to AK1\mathbf A^1_K has finite presentation: the coordinate rings are finite type over KK and Noetherian by [StacksAlg] (Tag 00FO). At the generic point its local map is K(t)→FK(t)\to F. This field map is flat by the ideal criterion [StacksAlg] (Tag 00HD), its maximal ideals are zero, and its residue extension is finite separable by hypothesis. The criterion [StacksAlg] (Tag 02GU) therefore makes the map étale at the generic point. Shrink to that algebraic étale open. Proposition 2.33.24 and Theorem 2.35.2 give an actual analytic étale germ at vv.

For any nonzero polynomial Q∈K[T]Q\in K[T], divide by a coefficient of maximal absolute value and reduce at vv. Transcendence of tˉ\bar t gives ∣Q(t)(v)∣=max⁡i∣Qi∣.|Q(t)(v)|=\max_i|Q_i|. Consequently this germ sends vv to the unit Gauss point, and its actual completed-residue-field map is Ht→LH_t\to L. Condition (3) and the finite separable residual extension in (2) are exactly the hypotheses of Proposition 4.52.3. It supplies an integral étale Gauss chart whose generic fibre is an actual affinoid domain of CanC^{\mathrm{an}} and whose component point is vv. The equivalence in Theorem 4.51.3, using the assumed existence of one original-field model, realizes vv as a component vertex of such a model. Every norm and every field embedding in the degree identity is the one induced by the original point vv. ◻

4.54 Unit degrees and original-field Gauss charts

All valued fields in this section are complete for their specified nontrivial real-valued non-Archimedean absolute values. For a valued field HH, write H∘H^\circ, H∘∘H^{\circ\circ} and H~\widetilde H for its valuation ring, maximal ideal and residue field. The trace is the trace of the finite-dimensional multiplication operator.

Definition 4.54.1. (Immediate extensions). A valued extension L/HL/H is immediate if ∣L×∣=∣H×∣|L^\times|=|H^\times| and the residue inclusion H~→L~\widetilde H\to\widetilde L is an isomorphism.

Lemma 4.54.2. NAG-002433 (A contractive normalized trace). Let L/HL/H be a finite extension of degree nn, with the extending absolute value. Then ∣Tr⁡L/H(x)∣≤∣x∣(x∈L).|\operatorname{Tr}_{L/H}(x)|\leq |x| \qquad(x\in L). If ∣n∣=1|n|=1, the map P=n−1Tr⁡L/H:L⟶HP=n^{-1}\operatorname{Tr}_{L/H}:L\longrightarrow H is an HH-linear retraction of norm at most one. In particular, a finite immediate extension of degree nn with ∣n∣=1|n|=1 is trivial.

Proof. Fix x∈Lx\in L, and let Q(T)=Td+ad−1Td−1+⋯+a0Q(T)=T^d+a_{d-1}T^{d-1}+\cdots+a_0 be its minimal polynomial over HH. Choose an algebraic closure of LL by [StacksAlg] (Tag 09GT) and let M/HM/H be the finite extension generated by the roots of QQ. Give MM its extending absolute value using Lemma 1.12.1. Every root xix_i of QQ, counted with its multiplicity, has absolute value ∣x∣|x|: the HH-isomorphism H(x)→H(xi)H(x)\to H(x_i) is isometric by uniqueness of the extending absolute value. Thus ∣ad−1∣=∣∑i=1dxi∣≤∣x∣.|a_{d-1}|=\left|\sum_{i=1}^d x_i\right| \leq |x|. Choose an H(x)H(x)-basis e1,…,eme_1,\ldots,e_m of LL. In the HH-basis xjeix^je_i, 0≤j<d0\leq j<d, multiplication by xx has mm companion-matrix blocks, each of trace −ad−1-a_{d-1}. Consequently Tr⁡L/H(x)=−mad−1.\operatorname{Tr}_{L/H}(x)=-m a_{d-1}. The ultrametric inequality gives ∣m∣≤1|m|\leq1, proving the bound. For h∈Hh\in H, its multiplication matrix is hh times the identity, so Tr⁡L/H(h)=nh\operatorname{Tr}_{L/H}(h)=nh. This proves the assertion about PP when ∣n∣=1|n|=1.

Suppose now that L/HL/H is immediate and nontrivial. The retraction PP has a nonzero kernel; take 0≠x∈ker⁡P0\ne x\in\ker P. Equality of value groups supplies c∈H×c\in H^\times with ∣c∣=∣x∣|c|=|x|. Since the residue fields agree, choose b∈H∘b\in H^\circ with the same residue as x/cx/c. Then a=cba=cb satisfies ∣x−a∣<∣x∣=∣a∣.|x-a|<|x|=|a|. On the other hand, P(x−a)=−aP(x-a)=-a, so contractivity gives ∣a∣≤∣x−a∣|a|\leq|x-a|, a contradiction. ◻

Proposition 4.54.3. NAG-002434 (Unchanged value group and a unit degree). Let L/HL/H be a finite extension of degree nn. Suppose that ∣L×∣=∣H×∣,∣n∣=1,L~/H~ is separable.|L^\times|=|H^\times|,\qquad |n|=1, \qquad \widetilde L/\widetilde H\text{ is separable}. Then [L:H]=[L~:H~].[L:H]=[\widetilde L:\widetilde H]. More precisely, LL is generated by a root of a monic polynomial over H∘H^\circ with irreducible separable reduction. Lifts of any H~\widetilde H-basis of L~\widetilde L form an orthonormal HH-basis of LL and an H∘H^\circ-basis of L∘L^\circ.

Proof. The residue extension is finite. Indeed, lifts of any finite residue-independent list are linearly independent over HH: normalize a proposed nonzero relation by a coefficient of maximal absolute value and reduce. Thus its degree ff is at most nn.

Choose a primitive residue element bˉ\bar b by [StacksAlg] (Tag 030N). Lift its monic minimal polynomial to Q∈H∘[T]Q\in H^\circ[T], of degree ff. The simple-root construction in Lemma 4.15.1, applied to LL, gives a root b∈L∘b\in L^\circ with residue bˉ\bar b. The powers 1,b,…,bf−11,b,\ldots,b^{f-1} are residue-independent and hence linearly independent over HH. Since Q(b)=0Q(b)=0, the intermediate field E=H(b)E=H(b) has degree ff. The same normalization of a linear combination proves ∣∑i=0f−1cibi∣=max⁡i∣ci∣(ci∈H).\left|\sum_{i=0}^{f-1}c_i b^i\right|= \max_i|c_i| \qquad(c_i\in H). It follows that E∘=∑iH∘biE^\circ=\sum_i H^\circ b^i, E~=L~\widetilde E=\widetilde L, and ∣E×∣=∣H×∣|E^\times|=|H^\times|. The field EE is complete by Lemma 1.12.1. Consequently L/EL/E is immediate.

Put m=[L:E]m=[L:E], so n=mfn=mf. Both integers have absolute value at most one, and ∣n∣=1|n|=1 forces ∣m∣=1|m|=1. Lemma 4.54.2 now gives L=EL=E. This proves the degree equality and the monic presentation. For any other lifted residue basis, the reduction of its change-of-basis matrix is invertible. Its determinant is a unit of H∘H^\circ, so the inverse matrix is integral by the adjugate formula. The same reduction argument gives the asserted norm identity and the two basis assertions. ◻

Lemma 4.54.4. NAG-002435 (A separating rational Gauss germ). Let KK be as above, put k=K~k=\widetilde K, and let C/KC/K be a smooth projective geometrically connected curve. Suppose v∈Canv\in C^{\mathrm{an}} has residue field l=H(v)~l=\widetilde{\mathcal H(v)} separably generated of transcendence degree one over kk. For every separating transcendental element tˉ∈l\bar t\in l there is a rational function t∈K(C)t\in K(C), defined at vv, such that ∣t(v)∣=1|t(v)|=1, its residue is tˉ\bar t, and the induced map to the analytic affine line is étale on an ordinary neighbourhood of vv. Moreover K(C)/K(t)K(C)/K(t) is finite separable, vv maps to the unit Gauss point, and Ht⟶H(v)H_t\longrightarrow \mathcal H(v) is finite, where HtH_t is the unit Gauss completion of K(t)K(t). In particular l/k(tˉ)l/k(\bar t) is finite separable.

Proof. The scheme image of vv is the generic point. Otherwise it is a closed point on the curve, with residue field finite over KK by [StacksAlg] (Tag 00FV). Proposition 2.33.28 and Lemma 1.12.1 then make H(v)/K\mathcal H(v)/K finite. Its residue extension is algebraic by Lemma 3.22.2, a contradiction. The same analytification proposition identifies K(C)K(C) with a dense valued subfield of L=H(v)L=\mathcal H(v).

Choose t0∈K(C)t_0\in K(C) of norm one with residue tˉ\bar t. By [StacksAlg] (Tag 054L), a smooth affine chart at the generic point has a rational étale coordinate ss. Its invertible Jacobian subsystem expresses all differentials in terms of dsds, and gives ΩK(C)/K=K(C) ds\Omega_{K(C)/K}=K(C)\,ds. Fix 0≠π∈K∘∘0\ne\pi\in K^{\circ\circ}. For sufficiently large NN, t=t0+πNst=t_0+\pi^Ns still has residue tˉ\bar t. We can also require dt≠0dt\ne0: at most one of the distinct scalars πN\pi^N can make dt0+πNdsdt_0+\pi^Nds vanish.

The residue transcendence shows that tt is transcendental over KK. The finitely generated field F=K(C)F=K(C) is algebraic over K(t)K(t) and hence finite over it. The universal property of derivations gives ΩF/K(t)=ΩF/K/Fdt=0\Omega_{F/K(t)}=\Omega_{F/K}/Fdt=0. On an affine open where tt is regular, the associated map to the affine line is of finite presentation: present its algebra as a quotient of a polynomial algebra over K[T]K[T] and apply the Noetherian assertion [StacksAlg] (Tag 00FO) to its kernel. The generic local map is a field extension, flat by the ideal criterion [StacksAlg] (Tag 00HD). By parts (4), (1) and (6) of [StacksAlg] (Tag 02GU), the map is étale at the generic point and F/K(t)F/K(t) is finite separable. Its étale locus is an algebraic open containing the scheme image of vv. Proposition 2.33.24 and Theorem 2.35.2 give the asserted ordinary analytic neighbourhood.

For a nonzero polynomial Q∈K[T]Q\in K[T], divide by a coefficient of largest absolute value and reduce at vv. The nonzero reduced polynomial cannot vanish at the transcendental tˉ\bar t, so ∣Q(t)∣=max⁡i∣Qi∣.|Q(t)|=\max_i|Q_i|. This is the unit Gauss norm. Completion embeds HtH_t isometrically in LL. The image of F⊗K(t)Ht→LF\otimes_{K(t)}H_t\to L is a finite-dimensional HtH_t-subalgebra of a field. It is a field: multiplication by any nonzero element is an injective, hence surjective, endomorphism of that finite-dimensional vector space. By Lemma 1.12.1 it is complete and closed in LL. It contains the dense field FF, so it equals LL. Finally, residue-independent lifts are linearly independent over HtH_t, as in Proposition 4.54.3. Thus [l:k(tˉ)]≤[L:Ht]<∞[l:k(\bar t)]\leq[L:H_t]<\infty; separability was part of the choice of the separating element. ◻

Theorem 4.54.5. NAG-002436 (Numerical vertex recognition in residue characteristic zero). Let KK be a complete nontrivially real-valued field with residue characteristic zero, and let C/KC/K be a smooth projective geometrically connected curve. If v∈Canv\in C^{\mathrm{an}} satisfies ∣H(v)×∣=∣K×∣,H(v)~/K~ is separably generated of transcendence degree one,\begin{aligned} |\mathcal H(v)^\times|&=|K^\times|,\\ \widetilde{\mathcal H(v)}/\widetilde K &\text{ is separably generated of transcendence degree one}, \end{aligned} then vv is the unique component point of an actual integral étale Gauss chart in CanC^{\mathrm{an}} and has an unramified branch certificate.

If CC admits one semistable algebraic model, in the sense of Definition 4.45.3, over the original K∘K^\circ, the two displayed conditions characterize the points which occur as component vertices of such models. In this existence class, a finite nonempty set satisfying those conditions is realized exactly with a specified unramified marking datum precisely when its entire geometric inverse image satisfies the marked semistable-vertex condition of Definition 4.48.3.

For arbitrary residue characteristic, the chart and branch-certificate conclusions, and realization under original-model existence, hold at a point satisfying the displayed conditions if one separating rational Gauss germ supplied by Lemma 4.54.4 has completed degree n=[H(v):Ht]n=[\mathcal H(v):H_t] with ∣n∣=1|n|=1.

Proof. Choose the rational Gauss germ by Lemma 4.54.4. Put L=H(v)L=\mathcal H(v) and l=L~l=\widetilde L. The Gauss calculation in Proposition 4.50.2 gives ∣Ht×∣=∣K×∣,Ht~=k(tˉ).|H_t^\times|=|K^\times|,\qquad \widetilde{H_t}=k(\bar t). Thus L/HtL/H_t has unchanged value group and finite separable residue extension. In residue characteristic zero every positive integer has absolute value one. Proposition 4.54.3 therefore gives [L:Ht]=[l:k(tˉ)].[L:H_t]=[l:k(\bar t)]. In arbitrary residue characteristic the stated hypothesis ∣n∣=1|n|=1 gives exactly the same conclusion by that proposition. Proposition 4.52.3 supplies the actual integral chart, and Theorem 4.50.4 supplies its branch certificate.

Under original-model existence, Theorem 4.48.2 realizes the point. The converse follows from Proposition 4.45.2. For a finite nonempty set, the pointwise certificates and the entire geometric marked-complement condition are precisely the hypotheses and conclusions of Theorem 4.48.4. All coordinates, completions and traces use the original absolute value; the argument does not require a discrete value group. ◻

4.55 An immediate extension and a nonrealizable curve point

For a complete valued field FF, write GF(T)G_F(T) for the completion of F(T)F(T) in its unit Gauss absolute value. All field embeddings below retain the specified absolute values.

Lemma 4.55.1. NAG-002438 (Finite constant extensions of a Gauss field). Let L/KL/K be a finite extension of complete real-valued fields. The canonical map GK(T)⊗KL⟶GL(T)G_K(T)\otimes_K L\longrightarrow G_L(T) is an isomorphism of fields and finite-dimensional normed GK(T)G_K(T)-algebras. In particular [GL(T):GK(T)]=[L:K].[G_L(T):G_K(T)]=[L:K]. No separability or discrete-valuation hypothesis is required.

Proof. Choose a KK-basis e1,…,ene_1,\ldots,e_n of LL. Lemma 1.12.1 gives constants c,C>0c,C>0 with cmax⁡i∣ai∣≤∣∑iaiei∣≤Cmax⁡i∣ai∣(ai∈K).c\max_i|a_i|\leq\left|\sum_i a_i e_i\right| \leq C\max_i|a_i| \qquad(a_i\in K). Apply these inequalities coefficient by coefficient to polynomials Pi∈K[T]P_i\in K[T] and then take the largest coefficient norm. The Gauss norms satisfy cmax⁡i∥Pi∥≤∥∑ieiPi∥≤Cmax⁡i∥Pi∥.c\max_i\|P_i\|\leq\left\|\sum_i e_iP_i\right\| \leq C\max_i\|P_i\|. The Gauss norm is multiplicative by Proposition 1.2.2. Division by a common nonzero denominator in K[T]K[T] therefore gives the same inequalities for any Pi∈K(T)P_i\in K(T). They also prove that the eie_i are independent over K(T)K(T). Their span is a finite-dimensional subalgebra of the field L(T)L(T), so it is a field: multiplication by a nonzero element is an injective endomorphism and hence surjective. It contains LL and TT, and is consequently all of L(T)L(T).

The displayed equivalent norms identify the completion of this vector space with ⨁iGK(T)ei\bigoplus_iG_K(T)e_i. Multiplication agrees on the dense subfield L(T)L(T) and extends continuously, since the Gauss norm is multiplicative. By definition its completion is GL(T)G_L(T). This proves the asserted map is an isomorphism and computes its degree. ◻

Proposition 4.55.2. NAG-002439 (An explicit immediate extension of degree pp). Fix a prime pp, a real number 0<ρ<10<\rho<1, and symbols q0,q1,…q_0,q_1,\ldots with qm+1p=qmq_{m+1}^p=q_m. Put Km=Fp((qm)),∣qm∣=ρ1/pm,K=⋃m≥0Km^,q=q0.K_m=\mathbf F_p((q_m)),\qquad |q_m|=\rho^{1/p^m}, \qquad K=\widehat{\bigcup_{m\geq0}K_m},\qquad q=q_0. Then KK is a complete field with a nondiscrete real-valued absolute value and ∣K×∣=ρZ[1/p],K~=Fp.|K^\times|=\rho^{\mathbf Z[1/p]},\qquad \widetilde K=\mathbf F_p. The polynomial P(S)=Sp−S−q−1P(S)=S^p-S-q^{-1} is irreducible over KK. For a root α\alpha and L=K(α)L=K(\alpha), the unique extending absolute value makes L/KL/K a degree-pp immediate extension.

Proof. In KmK_m, addition and multiplication are coefficientwise Laurent series operations. Every nonzero series has an invertible lowest term, and its inverse is obtained by a geometric series after factoring that term. Lowest nonzero terms prove multiplicativity of the indicated absolute value. A Cauchy sequence has eventually constant coefficients in every bounded exponent range and a common lower bound for its nonzero exponents, which proves completeness. Substitution qm=qm+1pq_m=q_{m+1}^p gives compatible isometric embeddings. Their union has precisely the displayed value group and residue field. Completion retains both by Lemma 3.22.2.

Let am=∑i=1mqi−1,a0=0.a_m=\sum_{i=1}^m q_i^{-1},\qquad a_0=0. Then P(am)=−qm−1P(a_m)=-q_m^{-1}. For b∈Kmb\in K_m, write b=am+cb=a_m+c. If ∣c∣≤1|c|\leq1, the term −qm−1-q_m^{-1} strictly dominates cp−cc^p-c. If ∣c∣>1|c|>1, its additive valuation is a negative integer multiple of 1/pm1/p^m, so v(cp−c)=p v(c)≤−p/pm<−1/pm.v(c^p-c)=p\,v(c)\leq-p/p^m<-1/p^m. Here v(q)=1v(q)=1 is used only to name the exponents; the absolute value remains ∣q∣=ρ|q|=\rho. It follows in either case that ∣P(b)∣≥ρ−1/pm>1.|P(b)|\geq\rho^{-1/p^m}>1. If PP had a root β∈K\beta\in K, density would give b∈Kmb\in K_m with ∣b−β∣<1|b-\beta|<1. But then P(b)=(b−β)p−(b−β)P(b)=(b-\beta)^p-(b-\beta) would have norm less than one, a contradiction.

Choose a root α\alpha in an algebraic closure supplied by [StacksAlg] (Tag 09GT). All roots of PP are the distinct elements α+c\alpha+c, c∈Fpc\in\mathbf F_p. If the minimal polynomial of α\alpha had degree dd with 0<d<p0<d<p, its next-to-leading coefficient would be −dα−∑i=1dci∈K-d\alpha-\sum_{i=1}^d c_i\in K for some ci∈Fpc_i\in\mathbf F_p. Since d≠0d\ne0 in Fp\mathbf F_p, this would give α∈K\alpha\in K. Thus PP is irreducible of degree pp. Equip LL with its unique complete extending absolute value by Lemma 1.12.1.

Write βm=α−am\beta_m=\alpha-a_m. Its equation is βmp−βm=qm−1.\beta_m^p-\beta_m=q_m^{-1}. The right side has norm greater than one. Hence ∣βm∣=ρ−1/pm+1|\beta_m|=\rho^{-1/p^{m+1}}. The norm-one element ym=qm+1βmy_m=q_{m+1}\beta_m satisfies ymp−qm+1p−1ym=1.y_m^p-q_{m+1}^{p-1}y_m=1. Its residue therefore satisfies yˉmp=1\bar y_m^p=1, and equals one in the residue field of characteristic pp.

Take 0≠x∈L0\ne x\in L. Write it in the basis 1,α,…,αp−11,\alpha,\ldots,\alpha^{p-1}. Approximate its finitely many coefficients by elements of one KmK_m, obtaining xm=∑j<pdjαjx_m=\sum_{j<p}d_j\alpha^j with dj∈Kmd_j\in K_m and ∣x−xm∣<∣x∣|x-x_m|<|x|. Expanding α=am+βm\alpha=a_m+\beta_m writes xm=∑j<pbjβmjx_m=\sum_{j<p}b_j\beta_m^j with bj∈Kmb_j\in K_m. The nonzero terms have pairwise different valuations: v(bjβmj)=v(bj)−j/pm+1,v(bj)∈p−mZ.v(b_j\beta_m^j)=v(b_j)-j/p^{m+1}, \qquad v(b_j)\in p^{-m}\mathbf Z. Indeed their numerators modulo pp distinguish the indices 0≤j<p0\leq j<p. Let jj index the unique term of largest norm, and put h=bjqm+1−j∈K×h=b_jq_{m+1}^{-j}\in K^\times. Then ∣h∣=∣xm∣=∣x∣|h|=|x_m|=|x|, while the residue of xm/hx_m/h is yˉmj=1\bar y_m^j=1. The strict approximation gives the same residue for x/hx/h. Every nonzero element of LL consequently has a value already in KK, and every norm-one element has residue in Fp\mathbf F_p. This proves immediacy. ◻

Theorem 4.55.3. NAG-002440 (The weak numerical conditions do not characterize vertices). Use KK, qq, LL and α\alpha from Proposition 4.55.2. Let J=GL(Y)J=G_L(Y) and put X=α+qYX=\alpha+qY in JJ. Restriction of this absolute value to K(X)K(X) defines a point v∈(PK1)anv\in(\mathbf P^1_K)^{\mathrm{an}} satisfying H(v)=GL(Y),∣H(v)×∣=∣K×∣,H(v)~=Fp(y).\begin{aligned} \mathcal H(v)&=G_L(Y),\\ |\mathcal H(v)^\times|&=|K^\times|,\qquad \widetilde{\mathcal H(v)}=\mathbf F_p(y). \end{aligned} Nevertheless vv is not a component vertex of any semistable algebraic model over the original K∘K^\circ in Definition 4.45.3.

The actual rational Gauss coordinate u=Xp−X−q−1qu=\frac{X^p-X-q^{-1}}q has ∣u(v)∣=1|u(v)|=1 and residue −y-y, and K(X)/K(u)K(X)/K(u) is finite separable. Its completed extension satisfies [H(v):GK(u)]=p,[H(v)~:Fp(uˉ)]=1.[\mathcal H(v):G_K(u)]=p, \qquad [\widetilde{\mathcal H(v)}:\mathbf F_p(\bar u)]=1. Thus the obstruction occurs on a curve which already has the smooth original model PK∘1\mathbf P^1_{K^\circ}.

Proof. The element XX is transcendental over KK, since YY is transcendental over LL. Its induced multiplicative seminorm on K[X]K[X] gives an affine analytic point by Theorem 2.33.23, and hence a point of the analytic projective line by Proposition 2.33.24. Proposition 2.33.28 identifies its completed residue field with the closure of K(X)K(X) in JJ. Set b=Xp−X−q−1=qpYp−qY.b=X^p-X-q^{-1}=q^pY^p-qY. The Gauss norm gives ∣b∣=∣q∣=ρ<1|b|=|q|=\rho<1. The series z=∑i≥0bpiz=\sum_{i\geq0}b^{p^i} converges in H(v)\mathcal H(v), satisfies zp−z=−bz^p-z=-b, and has norm at most ρ\rho. Hence X+zX+z is a root of PP. Its difference from α\alpha has norm at most ρ<1\rho<1, whereas every other root α+c\alpha+c, 0≠c∈Fp0\ne c\in\mathbf F_p, is at distance one from α\alpha. Therefore X+z=αX+z=\alpha. It follows that H(v)\mathcal H(v) contains LL and Y=(X−α)/qY=(X-\alpha)/q. Since L(Y)L(Y) is dense in JJ, it equals JJ. The unit Gauss calculation of Proposition 4.50.2, applied over LL, now gives the displayed value group and residue field.

We have u=qp−1Yp−Yu=q^{p-1}Y^p-Y, so ∣u∣=1|u|=1 and uˉ=−y\bar u=-y. Reduction of a normalized nonzero polynomial in uu shows that its induced norm is the unit Gauss norm, over both KK and LL. In GL(u)G_L(u) iterate Y0=−u,Yr+1=qp−1Yrp−u.Y_0=-u,\qquad Y_{r+1}=q^{p-1}Y_r^p-u. All iterates have norm at most one. Their successive differences decrease to zero, since the map has Lipschitz constant at most ∣q∣p−1<1|q|^{p-1}<1 on that ball. The element Y∈JY\in J is its fixed point, and the same estimate shows Yr→YY_r\to Y in JJ. The complete subfield GL(u)G_L(u) is closed, so it contains YY. Consequently J=GL(u)J=G_L(u), and Lemma 4.55.1 gives [J:GK(u)]=[L:K]=p[J:G_K(u)]=[L:K]=p. The residual degree is one because uˉ=−y\bar u=-y. The equation Xp−X−q−1−qu=0X^p-X-q^{-1}-qu=0 makes K(X)/K(u)K(X)/K(u) finite, and its derivative −1-1 makes the extension separable.

It remains to exclude all original models, not just this particular coordinate. Choose the comparison field Ω\Omega of Lemma 4.45.1, using an algebraic closure containing LL. For every c∈Fpc\in\mathbf F_p let wcw_c be the Gauss point over Ω\Omega of the disc ∣X−(α+c)∣≤∣q∣.|X-(\alpha+c)|\leq|q|. The map α↦α+c\alpha\mapsto\alpha+c defines a KK-automorphism of LL, and is isometric by Lemma 1.12.1. Applied coefficientwise to any polynomial in K[X]K[X], it shows that every wcw_c restricts to vv. These points are distinct: at wcw_c, the function X−(α+c)X-(\alpha+c) has norm ∣q∣|q|, while at wc′w_{c'} for c′≠cc'\ne c it has norm one.

Suppose vv were a model component vertex, with special component EE. Proposition 4.45.2 gives k(E)=Fp(y)k(E)=\mathbf F_p(y) and identifies the entire geometric inverse image of vv with the component generic points above EE. That set has one element. Indeed, Fp(y)⊗FpF‾p=F‾p(y)\mathbf F_p(y)\otimes_{\mathbf F_p} \overline{\mathbf F}_p =\overline{\mathbf F}_p(y) is a field. To verify the equality, every finite coefficient extension k′/Fpk'/\mathbf F_p stays linearly independent over Fp(y)\mathbf F_p(y), by clearing polynomial denominators. The tensor with k′k' is therefore a finite-dimensional domain and hence a field; it contains the inverses of all nonzero polynomials with coefficients in k′k'. Taking the union over these finite coefficient fields proves the equality. The comparison-field residue is F‾p\overline{\mathbf F}_p by Lemma 4.45.1. The pp distinct points wcw_c above vv contradict this singleton-fibre conclusion. Thus no such original model can have vv as a component vertex. Finally, the standard model PK∘1\mathbf P^1_{K^\circ} has its two polynomial affine charts, flat by the coefficientwise ideal criterion [StacksAlg] (Tag 00HD) and smooth with their displayed coordinates. It is projective and its special fibre is the geometrically integral projective line. It belongs to the stated model category. ◻

4.56 Semistable models after finite unramified extension

Let KK be a complete nontrivially real-valued field with its specified absolute value, and put O=K∘O=K^\circ and k=K~k=\widetilde K. Finite unramified extensions mean the lifts of finite separable residue extensions constructed in Lemma 4.15.2. Their absolute values extend the original one. We use the projective flat finitely presented nodal models of Definition 4.45.3 and the unramified marking data of Definition 4.47.1.

Proposition 4.56.1. NAG-002442 (Effective descent of an equivariant unramified model). Let L/KL/K be the finite unramified Galois lift of a finite Galois residue extension, put O′=L∘O'=L^\circ, and write G=Gal⁡(L/K)G=\operatorname{Gal}(L/K). Let C/KC/K be smooth, projective and geometrically connected, and let DD be an unramified marking datum on CC.

Suppose a semistable algebraic model Y/O′\mathcal Y/O' of CLC_L, marked by DO′D_{O'}, has a semilinear GG-action whose generic action is the canonical coefficient action on CLC_L and whose action on the marking is the canonical action on DO′D_{O'}. Then there is a marked semistable algebraic model X/O\mathcal X/O of (C,D)(C,D) and a compatible GG-equivariant isomorphism X×OO′≃Y.\mathcal X\times_O O'\simeq\mathcal Y. The projection Y→X\mathcal Y\to\mathcal X is finite étale of constant rank [L:K][L:K]. Descent is unique up to unique marked generic-identity isomorphism. Marked generic-identity morphisms between two such equivariant models descend uniquely if and only if they are GG-equivariant.

Proof. Lemma 4.43.1 gives an actual finite free étale and faithfully flat extension O′/OO'/O, of rank d=[L:K]d=[L:K], and the isomorphism O′⊗OO′≃∏g∈GO′,a⊗b⟼(a g(b))g.O'\otimes_O O'\simeq\prod_{g\in G}O',\qquad a\otimes b\longmapsto(a\,g(b))_g. Under this decomposition, the semilinear isomorphisms of Y\mathcal Y identify its two pullbacks to Spec⁡(O′⊗OO′)\operatorname{Spec}(O'\otimes_O O'). On the triple tensor product the cocycle condition is the group law. They therefore constitute a descent datum on the actual scheme Y\mathcal Y. Its structure morphism over O′O' is projective, so [StacksAlg] (Tag 0CCJ) makes this datum effective. We obtain an OO-scheme X\mathcal X and the displayed isomorphism, respecting the datum.

Fully faithful descent for schemes, [StacksAlg] (Tag 02W0), descends the specified generic isomorphism with CLC_L and its inverse to an isomorphism XK≃C\mathcal X_K\simeq C. It also descends the equivariant map DO′→YD_{O'}\to\mathcal Y to a map D→XD\to\mathcal X. The latter is a closed immersion by [StacksAlg] (Tag 02L6). The morphism X→Spec⁡O\mathcal X\to\operatorname{Spec}O is of finite presentation and proper by [StacksAlg] (Tags 02L0, 02L1). It is flat and at-worst-nodal of relative dimension one by [StacksAlg] (Tag 0C5E). These descent assertions apply over the valuation ring itself, without a Noetherian hypothesis.

The special fibre is geometrically connected and reduced. Indeed, for any extension k′/kk'/k choose a common field EE receiving k′k' and l=L~l=\widetilde L. Such a field is obtained from a maximal proper ideal of the nonzero finite-dimensional k′k'-algebra k′⊗klk'\otimes_k l; an ideal of maximal possible proper vector-space dimension is maximal. The base change Xk′×k′E\mathcal X_{k'}\times_{k'}E is the base change of Yl\mathcal Y_l to EE, and is connected and reduced. Projection onto Xk′\mathcal X_{k'} is surjective: at any point its fibre algebra is the nonzero finite-dimensional tensor of its residue field with EE, which has a field quotient by the same argument. Thus Xk′\mathcal X_{k'} is connected. Its affine coordinate rings inject into their scalar extensions to EE, so their nilpotents are zero. This proves both geometric assertions for every k′k'.

The marked points belong to the relative smooth locus. At a point zz of DD, choose a point z′z' of DO′D_{O'} above it. Differentials commute with this base change, as is seen from their universal property for derivations. Consequently (ΩX/O⊗κ(z))⊗κ(z)κ(z′)≃ΩY/O′⊗κ(z′).(\Omega_{\mathcal X/O}\otimes\kappa(z)) \otimes_{\kappa(z)}\kappa(z') \simeq\Omega_{\mathcal Y/O'}\otimes\kappa(z'). The vector space on the right has dimension one, because Y\mathcal Y is smooth of relative dimension one along its marking. The vector space before scalar extension therefore also has dimension one. The fibre of X/O\mathcal X/O has local dimension one at zz; its marking is finite over the base, so zz is closed in that fibre. Flatness and the differential criterion [StacksAlg] (Tag 01V9(4)) show that X/O\mathcal X/O is smooth at zz.

It remains to recover projectivity. The projection q:Y→Xq:\mathcal Y\to\mathcal X is the base change of Spec⁡O′→Spec⁡O\operatorname{Spec}O'\to\operatorname{Spec}O. It is finite free of rank dd on every affine open, surjective, and étale by [StacksAlg] (Tag 02GO). Because O′O' is affine and Y/O′\mathcal Y/O' is projective, Y\mathcal Y has an ample invertible sheaf by [StacksAlg] (Tag 0B45). The finite locally free case of [StacksAlg] (Tag 0BD4(1)) gives an ample invertible sheaf on X\mathcal X. Together with properness over the affine base, [StacksAlg] (Tag 0B45) now makes X/O\mathcal X/O projective. This norm argument uses the finite locally free rank dd; it involves no division by dd.

All the requirements of the stated model category are now satisfied. Finally [StacksAlg] (Tag 02W0) gives uniqueness of descent and full faithfulness on the indicated morphisms. The generic and marking compatibilities can be tested after the same faithful base change, so descend with the morphisms. Two generic-identity morphisms between the descended models are equal: their equalizer is closed by separatedness, and contains the schematically dense generic fibre of the flat source. On an affine open, this density is the injection of its torsion-free coordinate ring into its generic localization. ◻

Theorem 4.56.2. NAG-002443 (Unramified invariance of semistable-model existence). Let C/KC/K be smooth, projective and geometrically connected, let DD be an unramified marking datum, and let L/KL/K be any finite unramified extension. Then (C,D)(C,D) has a marked semistable algebraic model over the original OO if and only if (CL,DL∘)(C_L,D_{L^\circ}) has one over L∘L^\circ.

In the Galois case, given any unmarked semistable algebraic model Y/L∘\mathcal Y/L^\circ, one may choose the descended model so that its base change to L∘L^\circ dominates every Galois conjugate of Y\mathcal Y and carries the prescribed marking. No bound on the genus or assumption on the residue characteristic or discreteness of the valuation is required.

Proof. Base change of a projective model is projective by base change of its closed projective embedding. Finite presentation is preserved by the same finite algebra presentations, and nodality and flatness are preserved by [StacksAlg] (Tag 0C5B). The special fibre is the corresponding residue-field extension, as in Lemma 4.28.1; geometric connectedness and reducedness are preserved by their definitions. Closed immersions and finite étale markings are preserved by base change, the latter by [StacksAlg] (Tag 02GO); smoothness along them follows by base change of the standard smooth presentations in [StacksAlg] (Tag 01V9). This proves the forward implication.

First assume that L/KL/K is Galois, with group GG. Let Y/L∘\mathcal Y/L^\circ be a semistable algebraic model, initially without a marking. In (CL)an(C_L)^{\mathrm{an}} put S=⋃g∈GgV(Y).S=\bigcup_{g\in G}gV(\mathcal Y). This is a finite nonempty invariant set of component vertices of the conjugate models. Each has an unramified branch certificate by Theorem 4.48.2, applied over LL, where the model Y\mathcal Y already supplies the requisite existence hypothesis. Theorem 4.49.5 gives the least model MDO′(S)\mathcal M_{D_{O'}}(S) marked by DO′D_{O'}, where O′=L∘O'=L^\circ. Its vertex set contains each gV(Y)gV(\mathcal Y). Theorem 4.46.2 therefore gives actual generic-identity morphisms from it to all conjugate models.

The canonical semilinear action of GG on (CL,DO′)(C_L,D_{O'}) preserves SS. The equivariance assertion of the same least-hull theorem extends it uniquely to the actual algebraic model MDO′(S)\mathcal M_{D_{O'}}(S), with the group law and the specified action on the marking. Proposition 4.56.1 descends this model to the original OO. This proves both the reverse implication and the domination assertion in the Galois case.

For a general finite unramified L/KL/K, write l=L~l=\widetilde L. Choose a finite Galois extension l′/kl'/k containing ll. Concretely, take a primitive element of the finite separable extension l/kl/k by [StacksAlg] (Tag 030N), form the splitting field of its separable minimal polynomial by [StacksAlg] (Tag 09HU), and use [StacksAlg] (Tag 0EXM) for its Galois property. The field is finite because it is generated by finitely many algebraic roots. Lemma 4.18.1 supplies its finite unramified Galois lift L′/KL'/K.

There is an actual isometric KK-embedding L↪L′L\hookrightarrow L'. Indeed, the construction of Lemma 4.15.2 presents L=K(α)L=K(\alpha) by a monic minimal polynomial P∈O[T]P\in O[T] with separable irreducible reduction, and identifies αˉ∈l\bar\alpha\in l with its residue generator. Its image in l′l' is a simple root of Pˉ\bar P. The simple-root construction in Lemma 4.15.1, applied inside the complete field L′L', lifts it to a root α′∈(L′)∘\alpha'\in(L')^\circ of PP. Sending α\alpha to α′\alpha' embeds LL in L′L'; uniqueness of the extending absolute value in Lemma 1.12.1 makes this embedding isometric. It consequently embeds the actual valuation rings.

Base change the given model over L∘L^\circ to (L′)∘(L')^\circ, using the properties verified in the first paragraph. Apply the Galois case to L′/KL'/K. The resulting marked model is over the original valuation ring with the original generic curve and marking. All absolute values throughout are the specified extending values; none has been rescaled. ◻

4.57 Geometric components obstructing original-field models

Let KK be complete and nontrivially real-valued, with its fixed absolute value, and put O=K∘O=K^\circ and k=K~k=\widetilde K. Semistable algebraic models have the meaning of Definition 4.45.3. Choose the isometric algebraically closed comparison field Ω\Omega of Lemma 4.45.1; write k‾=Ω~\overline k=\widetilde\Omega. For a smooth projective geometrically connected curve C/KC/K, let p:CΩan→Canp:C_\Omega^{\mathrm{an}}\to C^{\mathrm{an}} be projection. A function field over k‾\overline k is called rational here if it is isomorphic over k‾\overline k to k‾(t)\overline k(t).

Proposition 4.57.1. NAG-002445 (A nonrational component cannot disappear). Let C/KC/K be a smooth projective geometrically connected curve. Suppose Y/Ω∘\mathcal Y/\Omega^\circ is a semistable algebraic model of CΩC_\Omega, and let ww be the component vertex of a special component EE whose function field k‾(E)\overline k(E) is not rational. Then every semistable algebraic model X/O\mathcal X/O of CC satisfies w∈V(XΩ∘),p(w)∈V(X).w\in V(\mathcal X_{\Omega^\circ}), \qquad p(w)\in V(\mathcal X). Consequently, if CC has such an original-field model, the point v=p(w)v=p(w) must satisfy ∣H(v)×∣=∣K×∣,H(v)~/k is separably generated of transcendence degree one.\begin{aligned} |\mathcal H(v)^\times|&=|K^\times|,\\ \widetilde{\mathcal H(v)}/k &\text{ is separably generated of transcendence degree one}. \end{aligned} Failure of either condition at vv rules out every original-field semistable algebraic model, including every such marked model.

Proof. Assume X\mathcal X exists. Its coefficient extension XΩ∘\mathcal X_{\Omega^\circ} is a semistable algebraic model of the same curve as Y\mathcal Y, by Lemma 4.28.1 and preservation of the nodal relative-curve condition under base change, [StacksAlg] (Tag 0C5B). Apply Theorem 4.47.5 over the complete field Ω\Omega, with empty marking. It supplies an actual semistable model T/Ω∘\mathcal T/\Omega^\circ and generic-identity morphisms T⟶Y,T⟶XΩ∘.\mathcal T\longrightarrow\mathcal Y, \qquad \mathcal T\longrightarrow\mathcal X_{\Omega^\circ}. The first morphism and Proposition 4.45.6 show that w∈V(T)w\in V(\mathcal T). Let FF be its special component in T\mathcal T. The actual residue-field identifications of Proposition 4.45.2, applied over Ω\Omega, give k‾(F)=H(w)~=k‾(E).\overline k(F)=\widetilde{\mathcal H(w)} =\overline k(E). If ww were absent from V(XΩ∘)V(\mathcal X_{\Omega^\circ}), the necessity of the geometric forest condition for the second morphism would make FF a smooth Pk‾1\mathbf P^1_{\overline k}. Its function field would be rational, contrary to the displayed equality. Hence ww is a vertex of the coefficient extension of X\mathcal X. The geometric vertex identification in Proposition 4.45.2 projects it to the original vertex v=p(w)v=p(w) and gives the two asserted arithmetic conditions. Forgetting a marking retains a semistable algebraic model, so the same obstruction applies to every prescribed marking. ◻

Theorem 4.57.2. NAG-002446 (A potential smooth model with an original-field obstruction). Assume char⁡k∉{2,3}\operatorname{char}k\notin\{2,3\}, and choose a∈Ka\in K with 0<∣a∣<10<|a|<1. Let CaC_a be the projective curve y2z=x3+az3.y^2z=x^3+a z^3. It is smooth and geometrically connected. If ∣a∣1/2∉∣K×∣or∣a∣1/3∉∣K×∣,|a|^{1/2}\notin|K^\times| \quad\text{or}\quad |a|^{1/3}\notin|K^\times|, then CaC_a has no semistable algebraic model over the original valuation ring OO.

Nevertheless, if b∈Ωb\in\Omega satisfies b6=ab^6=a and L=K(b)L=K(b), the curve has the smooth projective model E/L∘:Y2Z=X3+Z3,[x:y:z]=[b2X:b3Y:Z].\mathcal E/L^\circ:\quad Y^2Z=X^3+Z^3, \qquad [x:y:z]=[b^2X:b^3Y:Z]. The unique component vertex of EΩ∘\mathcal E_{\Omega^\circ} is fixed by every isometric KK-automorphism of Ω\Omega. In particular, for K=QpK=\mathbf Q_p, p≥5p\geq5, with the specified absolute value ∣p∣=p−1|p|=p^{-1}, the choice a=pa=p gives the stated nonexistence over Zp\mathbf Z_p.

Proof. We first verify the actual model and its special curve. The integers 22 and 33 are units in OO, and hence in L∘L^\circ. The two affine charts Z≠0Z\ne0 and Y≠0Y\ne0 cover E\mathcal E: the equation rules out a projective point with Y=Z=0Y=Z=0. Their coordinate rings are L∘[X,Y]/(Y2−X3−1),L∘[u,v]/(v−u3−v3).\begin{gathered} L^\circ[X,Y]/(Y^2-X^3-1),\\ L^\circ[u,v]/(v-u^3-v^3). \end{gathered} Each is free over the polynomial algebra in the first variable, since its equation is monic, up to a unit, in the second. They are therefore flat over L∘L^\circ. They have finite algebra presentations; the projective closed embedding makes the resulting morphism separated and quasi-compact, so it is finitely presented.

On the first chart, Y2−X3=1Y^2-X^3=1 makes D(X)D(X) and D(Y)D(Y) a cover; the corresponding partial derivatives −3X2-3X^2 and 2Y2Y are units. On the second chart, a point with v≠0v\ne0 belongs to the first chart, whereas a point with v=0v=0 has 1−3v21-3v^2 invertible. These are actual standard smooth presentations by [StacksAlg] (Tag 01V9(6)). Thus E/L∘\mathcal E/L^\circ is smooth of relative dimension one.

Over any extension field of L~\widetilde L, the polynomial X3+1X^3+1 is not a square in the rational function field in XX: its order at infinity is −3-3, whereas the order of a square is even. Therefore Y2−X3−1Y^2-X^3-1 is irreducible. Its quotient ring injects into the corresponding quadratic field over that rational function field, by its monic degree-two basis, so the affine chart is integral. The only point of its projective complement is [0:1:0][0:1:0]. Every fibre is smooth of relative dimension one; in particular this point is on a one-dimensional component and cannot be an additional component by itself. The affine chart is consequently dense, and the fibre is connected. It is geometrically reduced and regular by [StacksAlg] (Tag 056T). This verifies the exact semistable-model conditions, including geometric connectedness, over L∘L^\circ.

The same calculation with a≠0a\ne0 over any extension of KK proves that CaC_a is smooth and geometrically connected: the affine chart has equation y2=x3+ay^2=x^3+a, with partial derivatives 2y2y and −3x2-3x^2, which cannot both vanish on it; the unique point at infinity has a unit partial derivative in the chart y≠0y\ne0. Its affine polynomial is irreducible by the same odd-order calculation. The root bb exists in the chosen algebraically closed Ω\Omega. The extension L/KL/K has its actual degree [K(b):K]≤6[K(b):K]\leq6 and is complete with the extending absolute value by Lemma 1.12.1. The displayed coordinate change identifies the generic fibre of E\mathcal E with Ca,LC_{a,L}, and ∣b∣=∣a∣1/6|b|=|a|^{1/6} for this unchanged extending value.

Write EE for the special curve after extension to k‾\overline k. Its function field is not rational. Here is a direct verification. On the affine chart the differential ω=dX2Y\omega=\frac{dX}{2Y} extends across a zero of YY as dY/(3X2)dY/(3X^2), since 2Y dY=3X2 dX2Y\,dY=3X^2\,dX and X≠0X\ne0 there. Near infinity put u=X/Yu=X/Y and v=Z/Yv=Z/Y, so v=u3+v3v=u^3+v^3. The affine coordinates are u/vu/v and 1/v1/v; differentiating the equation gives (1−3v2) dv=3u2 du,ω=v du−u dv2v=−du1−3v2.(1-3v^2)\,dv=3u^2\,du,\qquad \omega=\frac{v\,du-u\,dv}{2v} =-\frac{du}{1-3v^2}. This is regular at infinity. The differential is nonzero: on the affine function field the derivation δ(X)=2Y\delta(X)=2Y, δ(Y)=3X2\delta(Y)=3X^2 respects the equation and evaluates ω\omega to 11.

If k‾(E)\overline k(E) were k‾(t)\overline k(t), [StacksAlg] (Tag 0BY1), applied to the smooth projective curve EE and Pk‾1\mathbf P^1_{\overline k}, would identify these curves. But a regular differential on P1\mathbf P^1 has the form f(t) dtf(t)\,dt on its affine line, with f∈k‾[t]f\in\overline k[t]; at s=t−1s=t^{-1} it is −f(s−1)s−2 ds-f(s^{-1})s^{-2}\,ds, which is regular at s=0s=0 only when f=0f=0. The nonzero ω\omega is a contradiction.

Complete EΩ∘\mathcal E_{\Omega^\circ} along any fixed 0≠π∈K∘∘0\ne\pi\in K^{\circ\circ}. Theorem 4.13.2 identifies its actual analytic generic fibre with Ca,ΩanC_{a,\Omega}^{\mathrm{an}}. Proposition 4.12.2 supplies the unique component vertex ww over the generic point of EE. Put v=p(w)v=p(w). The reduction identification in Proposition 4.24.2 shows that the affine functions XX and YY have norm one at ww, since both have nonzero residues at the generic point of EE. Hence the original rational functions x/zx/z and y/zy/z satisfy ∣(x/z)(v)∣=∣a∣1/3,∣(y/z)(v)∣=∣a∣1/2.|(x/z)(v)|=|a|^{1/3},\qquad |(y/z)(v)|=|a|^{1/2}. They are nonzero elements of H(v)\mathcal H(v), so either failure in the hypothesis makes its value group strictly larger than ∣K×∣|K^\times|. Proposition 4.57.1 now rules out every original-field semistable model.

Finally, let σ\sigma be an isometric KK-automorphism of Ω\Omega. Then σ(b)=ζb\sigma(b)=\zeta b with ζ6=1\zeta^6=1, and ∣ζ∣=1|\zeta|=1. The canonical semilinear action on Ca,ΩC_{a,\Omega} transforms the displayed integral coordinates by X⟼ζ−2X,Y⟼ζ−3Y,Z⟼Z.X\longmapsto\zeta^{-2}X, \qquad Y\longmapsto\zeta^{-3}Y, \qquad Z\longmapsto Z. These substitutions and their inverses preserve the equation over Ω∘\Omega^\circ, so give an actual semilinear automorphism of the smooth model. Its unique special component, and thus its component vertex ww, is preserved. For K=QpK=\mathbf Q_p and a=pa=p, the two displayed values are p−1/3p^{-1/3} and p−1/2p^{-1/2}, neither of which lies in ∣Qp×∣=pZ|\mathbf Q_p^\times|=p^{\mathbf Z}. Throughout, the absolute value was extended and never rescaled. ◻