D · Polarized towers with compatible cyclic actions

3 NAG Tags

A polarized isomorphism (X,L)→(Y,M)(X,\mathcal L)\to(Y,\mathcal M) is a pair (f,λ)(f,\lambda) consisting of an isomorphism of schemes and a specified isomorphism λ:f∗M≃L\lambda:f^*\mathcal M\simeq\mathcal L. Thus its line-bundle map is part of the datum.

Proposition D.0.1. NAG-002907 (Orbit tensor and its canonical linearization). Let a cyclic group G=⟨τ⟩G=\langle\tau\rangle of order ee act on a scheme XX, and let A\mathcal A be an invertible sheaf on XX, with no chosen linearization. The orbit tensor N=⨂j=0e−1(τj)∗A(6)\mathcal N=\bigotimes_{j=0}^{e-1}(\tau^j)^*\mathcal A \tag{6} has a canonical GG-linearization. If i:Z↪Xi:Z\hookrightarrow X is an invariant closed subscheme and GG acts identically on ZZ as a scheme, then this linearization acts identically on i∗Ni^*\mathcal N. If XX is separated over an affine scheme SS and A\mathcal A is relatively ample over SS, then so is N\mathcal N; the group action need not be SS-linear.

Proof. The factors of τ∗N\tau^*\mathcal N have indices 1,2,…,e1,2,\ldots,e. Using τe=id⁡\tau^e=\operatorname{id}, reorder them cyclically into indices 0,1,…,e−10,1,\ldots,e-1. This gives λ:τ∗N≃N\lambda:\tau^*\mathcal N\simeq\mathcal N. Iterating this construction ee times returns every factor to its original position. The ordinary tensor symmetry is the permutation of pure tensors, so the composite is the identity. The iterates of λ\lambda therefore satisfy the cyclic group law and define the claimed linearization.

On ZZ, the equality τj∘i=i\tau^j\circ i=i gives canonical identifications i∗(τj)∗A=i∗A,i∗N≃(i∗A)⊗e.(7)i^*(\tau^j)^*\mathcal A=i^*\mathcal A, \qquad i^*\mathcal N\simeq(i^*\mathcal A)^{\otimes e}. \tag{7} The restricted λ\lambda is the cyclic permutation of these identical line-bundle factors. On an open of ZZ where i∗Ai^*\mathcal A has a generator aa, it fixes a⊗ea^{\otimes e}. It fixes every coefficient as well, because τ∣Z\tau|_Z is the identity morphism. Hence it fixes every local section. This proves identity of the full module, including its behaviour over any nilpotent functions on ZZ. It also proves identity on the associated total line-bundle scheme.

For ampleness, Stacks Tag 01VK turns relative ampleness over the affine SS into ampleness on XX. Each automorphism τj\tau^j preserves ampleness, by Stacks Tag 0B3E. For two ample invertible sheaves H1,H2\mathcal H_1,\mathcal H_2 on XX, Stacks Tag 0DNK makes pr⁡1∗H1⊗pr⁡2∗H2\operatorname{pr}_1^*\mathcal H_1\otimes \operatorname{pr}_2^*\mathcal H_2 ample on X×SXX\times_S X. Here SS is quasi-separated because it is affine. Pulling back along the closed diagonal by Stacks Tag 0B3E shows that H1⊗H2\mathcal H_1\otimes\mathcal H_2 is ample on XX. Induction gives ampleness of (6), and Stacks Tag 01VK converts it back to relative ampleness over SS. ◻

Lemma D.0.2. NAG-002985 (Full faithfulness for polarized isomorphisms). Let RR be a ring, let π∈R\pi\in R, and suppose that the canonical map R→lim←⁡n≥1R/πnRR\to\varprojlim_{n\geq1}R/\pi^nR is an isomorphism. Write Rn=R/πnRR_n=R/\pi^nR. Let X,YX,Y be proper, flat, finitely presented RR-schemes, and let L,M\mathcal L,\mathcal M be relatively ample invertible sheaves on them. Reduction gives a bijection Isom⁡Rpol((X,L),(Y,M))→ ∼ lim←⁡nIsom⁡Rnpol((Xn,Ln),(Yn,Mn)).(1)\operatorname{Isom}^{\rm pol}_R((X,\mathcal L),(Y,\mathcal M)) \xrightarrow{\ \sim\ } \varprojlim_n\operatorname{Isom}^{\rm pol}_{R_n} ((X_n,\mathcal L_n),(Y_n,\mathcal M_n)). \tag{1} Here the inverse limit means compatible polarized isomorphisms with respect to the specified reduction maps. The analogous statement holds for polarized isomorphisms covering a fixed automorphism bb of Spec⁡R\operatorname{Spec}R that preserves the closed subscheme Spec⁡Rn\operatorname{Spec}R_n for every nn. No Noetherian hypothesis on RR is imposed.

Proof. Apply Stacks Tag 0D4F to each polarized scheme, allowing the identity thickening of its affine base, and enlarge the two bounds to a common d0≥1d_0\geq1. For every d≥d0d\geq d_0, the modules Pd=H0(X,L⊗d),Qd=H0(Y,M⊗d)P_d=H^0(X,\mathcal L^{\otimes d}),\qquad Q_d=H^0(Y,\mathcal M^{\otimes d}) are finite projective over RR. The arbitrary-base-change assertion in the proof of that lemma identifies their reductions with H0(Xn,Ln⊗d)H^0(X_n,\mathcal L_n^{\otimes d}) and H0(Yn,Mn⊗d)H^0(Y_n,\mathcal M_n^{\otimes d}). Moreover the graded algebras AX=R⊕⨁d≥d0Pd,AY=R⊕⨁d≥d0Qd(2)A_X=R\oplus\bigoplus_{d\geq d_0}P_d, \qquad A_Y=R\oplus\bigoplus_{d\geq d_0}Q_d \tag{2} functorially reconstruct the pairs as (Proj⁡AX,O(1))(\operatorname{Proj}A_X,\mathcal O(1)) and (Proj⁡AY,O(1))(\operatorname{Proj}A_Y,\mathcal O(1)).

For any finite projective RR-module PP, the canonical map P→lim←⁡nP/πnPP\to\varprojlim_n P/\pi^nP is an isomorphism. Indeed, write PP as the image of an idempotent matrix EE on RmR^m. The assertion becomes ERm=lim←⁡nE‾RnmE R^m=\varprojlim_n\overline E R_n^m, which follows from the assumed identity for RR. This is the finite-projective completeness argument in the proof of Stacks Tag 0D4B.

Let (fn,λn)(f_n,\lambda_n) be a compatible system on the right of (1). For every d≥d0d\geq d_0, its maps on sections are un,d:Qd/πnQd⟶Pd/πnPd,s⟼λn⊗d(fn∗s).(3)u_{n,d}:Q_d/\pi^nQ_d\longrightarrow P_d/\pi^nP_d, \qquad s\longmapsto\lambda_n^{\otimes d}(f_n^*s). \tag{3} Taking the inverse limit in each degree gives ud:Qd→Pdu_d:Q_d\to P_d. The same construction applied to the inverse polarized isomorphisms gives inverse maps. The udu_d preserve products: after reduction modulo every πn\pi^n the required equality holds by (3), and its values lie in the separated module Pd+d′P_{d+d'}. Together with the identity on degree zero, they therefore give a graded isomorphism AY≃AXA_Y\simeq A_X.

Functoriality of Proj, including the maps on all twists in Stacks Tag 01MY, gives an isomorphism of schemes and an isomorphism of their O(1)\mathcal O(1) sheaves. The canonical identifications from Stacks Tag 0D4F turn this into (f,λ)(f,\lambda) on the original pairs. Its reductions are exactly (fn,λn)(f_n,\lambda_n), by the functoriality of those identifications. Conversely, any polarized isomorphism inducing the prescribed reductions has the same maps (3), hence the same udu_d by completeness and separation. The same functorial reconstruction makes it equal to (f,λ)(f,\lambda). This proves both directions of (1).

The limit was taken degree by degree; it was not interchanged with an infinite direct sum. All degrees d≥d0d\geq d_0 were retained, so the reconstruction recovers the actual invertible sheaf and its map, not merely a selected tensor power. Finally, an isomorphism covering bb is an RR-isomorphism to the pullback pair b∗(Y,M)b^*(Y,\mathcal M). This twist commutes with reduction by the stated hypothesis on bb. Applying (1) to the twisted target proves the last assertion. ◻

Theorem D.0.3. NAG-002986 (Equivariant algebraization of a supplied nodal tower). Let FF be a complete nontrivially real-valued field, R=F∘R=F^\circ, mR=F∘∘\mathfrak m_R=F^{\circ\circ}, and kR=R/mRk_R=R/\mathfrak m_R. Let a cyclic group GG of order e≥1e\geq1 act isometrically on FF, and put bg=Spec⁡(g−1∣R)b_g=\operatorname{Spec}(g^{-1}|_R). Fix 0≠π∈mR0\ne\pi\in\mathfrak m_R, set Rn=R/πnRR_n=R/\pi^nR, and suppose given proper, flat, finitely presented RnR_n-schemes XnX_n with cartesian transition closed immersions in:Xn↪Xn+1i_n:X_n\hookrightarrow X_{n+1}. Suppose compatible semilinear GG-actions on these schemes are specified, covering the induced bg,nb_{g,n}. Assume that C=X1×R1kRC=X_1\times_{R_1}k_R is at-worst-nodal of pure dimension one and that the action on the entire scheme CC is the identity.

For any ample invertible sheaf AC\mathcal A_C on CC, one can choose compatible ample lifts An\mathcal A_n and form the orbit tensors Nn=⨂j=0e−1(τnj)∗An,(4)\mathcal N_n=\bigotimes_{j=0}^{e-1}(\tau_n^j)^*\mathcal A_n, \tag{4} where τn\tau_n is the action of a generator of GG. These form a compatible GG-linearized ample tower, with identity action on Nn∣C≃AC⊗e\mathcal N_n|_C\simeq\mathcal A_C^{\otimes e}. There exist a projective, flat, finitely presented nodal RR-scheme Y′Y' of relative dimension one, an ample invertible sheaf N′\mathcal N' on Y′Y', a semilinear GG-action on Y′Y', and a compatible linearization of N′\mathcal N', together with compatible equivariant polarized identifications (Y′,N′)×RRn≃(Xn,Nn).(5)(Y',\mathcal N')\times_R R_n \simeq (X_n,\mathcal N_n). \tag{5} The actions on both YkR′Y'_{k_R} and N′∣YkR′\mathcal N'|_{Y'_{k_R}} are the identity. For a fixed polarized tower (4), this algebraization is unique up to the unique compatible equivariant polarized isomorphism. Invertibility of ee is not required here.

Proof. The isometries preserve RR and every ideal πnR\pi^nR: membership in this ideal is the inequality v(a)≥nv(π)v(a)\geq n v(\pi). Hence all indicated actions on quotient bases exist. The rank-one completeness argument in B.1.3 gives R=lim←⁡RnR=\varprojlim R_n.

The exact result B.1.2 gives the compatible ample An\mathcal A_n, with transition isomorphisms φn:in∗An+1≃An\varphi_n:i_n^*\mathcal A_{n+1}\simeq\mathcal A_n and a specified restriction to AC\mathcal A_C. It also supplies at least one ample AC\mathcal A_C, if none was prescribed. Apply D.0.1 at each level. Its hypotheses hold because XnX_n is separated over the affine base Spec⁡Rn\operatorname{Spec}R_n. It gives ampleness of (4), its canonical permutation linearization, and identity on the full restricted line bundle on CC.

We check the extra tower compatibility explicitly. Equivariance of ini_n gives canonical identifications in∗(τn+1j)∗An+1≃(τnj)∗in∗An+1.i_n^*(\tau_{n+1}^j)^*\mathcal A_{n+1} \simeq(\tau_n^j)^*i_n^*\mathcal A_{n+1}. Tensor the maps (τnj)∗φn(\tau_n^j)^*\varphi_n over all jj to obtain νn:in∗Nn+1≃Nn\nu_n:i_n^*\mathcal N_{n+1}\simeq\mathcal N_n. Applying the cyclic permutation linearization before or after this tensor map permutes the same indexed factors and applies the same maps on them. Thus νn\nu_n is equivariant. Composing these maps gives compatibility at all levels. Their restrictions to CC identify (4) with AC⊗e\mathcal A_C^{\otimes e}, with the identity action already given by the orbit-tensor result.

Now apply B.1.3 to the ordinary polarized tower (Xn,Nn,νn)(X_n,\mathcal N_n,\nu_n). It gives the projective flat finitely presented pair (Y′,N′)(Y',\mathcal N') and the underlying polarized identifications (5). The nodal hypothesis on CC gives nodality of relative dimension one. At this stage no action on Y′Y' has been asserted.

Let bb be the base automorphism corresponding to the chosen generator. Via (5), its tower action and linearization are compatible polarized RnR_n-isomorphisms from (Y′,N′)Rn(Y',\mathcal N')_{R_n} to its bnb_n-twist. Both algebraic pairs (Y′,N′)(Y',\mathcal N') and b∗(Y′,N′)b^*(Y',\mathcal N') satisfy Lemma D.0.2. That lemma gives a unique polarized RR-isomorphism between them. Composing its scheme map with the first projection from the twist gives a semilinear automorphism τ:Y′→Y′\tau:Y'\to Y' covering bb, and its line-bundle map is λ:τ∗N′≃N′\lambda:\tau^*\mathcal N'\simeq\mathcal N'.

The ee-fold composite of this polarized semilinear automorphism covers be=id⁡b^e=\operatorname{id}. At every level it is the identity polarized automorphism, by the given action and its linearization. Injectivity in Lemma D.0.2 therefore makes it the identity on the algebraic pair. The iterates of (τ,λ)(\tau,\lambda) give the required GG-action and linearization, and (5) is equivariant. Restricting (5) for n=1n=1 further to CC proves the full residual identities. Finally, the same lemma gives a unique polarized comparison of any two algebraizations with (5). Its reductions commute with the specified actions, so injectivity makes the comparison equivariant. ◻