Bogomolov–Tian–Todorov theorems via crystalline periods of differential forms
Abstract.
We prove Bogomolov–Tian–Todorov theorems for smooth proper varieties over perfect fields of positive characteristic using differential forms whose contractions jointly detect obstructions. Numerical Hodge–de Rham degeneration and crystalline torsion-freeness one degree above each form ensure unobstructed deformations over the Witt vectors and formal smoothness of forgetting the forms. This refines Brantner–Taelman’s Calabi–Yau theorem and gives a symplectic criterion involving only degree three. We also give criteria for smooth projective lifts and symplectic examples with degree-three crystalline torsions.
Key words and phrases:
crystalline cohomology, deformation theory, differential forms, symplectic varieties, period maps, liftings2020 Mathematics Subject Classification:
Primary 14D15; Secondary 14F30, 14J32, 14J421. Introduction
The Bogomolov–Tian–Todorov theorem asserts that a compact Kähler manifold with trivial canonical bundle has unobstructed deformations, even though the obstruction space need not vanish. Contraction with a volume form makes obstructions visible in de Rham cohomology, where Hodge theory forces them to vanish. In positive characteristic this argument meets an essential obstruction: a closed form determines a de Rham class on the special fibre, but this class need not lift across a mixed-characteristic extension. Non-liftable varieties with trivial canonical bundle show that additional hypotheses are necessary [15, 9].
Brantner–Taelman resolve this difficulty for Calabi–Yau varieties whose Hodge–de Rham spectral sequence degenerates and whose crystalline cohomology is torsion-free [8, Theorem A]. Their animated crystalline period map and divided-power bootstrapping provide the framework for this paper. We prove a criterion whose hypotheses are tailored to the forms that detect obstructions. For a two-form the relevant cohomological degree is three, regardless of the dimension; for a volume form it is the middle degree; and several forms may detect obstructions jointly even when none does so individually. All conditions are imposed on the special fibre, and none presupposes a lift, even for .
We also construct a smooth projective symplectic fourfold in characteristic three whose Hodge–de Rham spectral sequence degenerates at and whose mixed-characteristic deformations are obstructed, despite admitting a projective symplectic lift over the Witt vectors (Theorem E).
1.1. The main criterion
Let be a perfect field of characteristic , and put . Throughout the paper, is smooth, proper, and geometrically connected, of dimension . We write for the functor of isomorphism classes of deformations of over local Artinian -algebras with residue field , and similarly for other objects over .
For an integer , write for the numerical equality
which holds for every if the Hodge–de Rham spectral sequence of degenerates at . Fix differential forms with and . Contraction with these forms defines the joint contraction map on obstructions
| (1.1) |
When the forms are closed, a deformation of is a deformation of together with closed relative lifts of all the forms.
Theorem A.
Assume that (1.1) is injective and that, for every , holds and is torsion-free.
-
(i)
The deformation functors of and of are unobstructed over .
-
(ii)
The morphism forgetting the forms is formally smooth on discrete Artinian -algebras.
Formal smoothness is meant in the relative sense: given a surjection , a deformation with forms over , and a marked lift to of its underlying variety, the forms lift to this given variety. In particular, every Artinian deformation of carries lifts of the forms . Theorem A holds without geometric connectedness, which we assume throughout only for the sake of the volume-form and symplectic applications.
1.2. Volume and symplectic forms
A nowhere-vanishing top form induces an isomorphism , and integral crystalline duality shows that is torsion-free if and only if is. Applying Theorem A gives the following degree-wise criteria for unobstructedness.
Corollary B.
Suppose that , that holds, and that is torsion-free. Fix a nowhere-vanishing top form . Then the deformation functors of and of are unobstructed over , and forgetting the volume form is formally smooth on discrete Artinian bases. 111When preparing this paper, we noticed that Takamatsu also obtain Corollary B (see [28, Proposition 2.11]) with different methods.
Corollary C.
Let have invertible contraction map . Assume that holds and that is torsion-free. Then is closed, the deformation functors of and of are unobstructed over , and forgetting the form is formally smooth on discrete Artinian bases.
Theorem A also applies beyond these two settings: for instance, Example 6.11 treats a product of plane quartics, whose canonical bundle is ample and whose obstruction space is nonzero.
1.3. Projective lifts and algebraization
To obtain a projective -scheme from compatible formal lifts, one must also extend an ample line bundle. For a line bundle on , put and define
| (1.2) |
Theorem D.
Assume the hypotheses of Theorem A, and let be an ample line bundle on for which (1.2) is surjective. Then the deformation functors of and are unobstructed over . In particular, there is a smooth projective -scheme lifting , together with a relatively ample lift of and closed relative lifts of all the .
The proof in Section 7.2 varies a geometric lift by to cancel the obstruction in , and then lifts the prescribed forms to the resulting variety. When , every compatible formal lift algebraizes with the chosen (Proposition 7.1). Section 7 also treats prescribed formal lifts, using a crystalline Chern-class criterion independent of Theorem A; in particular, every smooth proper formal lift algebraizes projectively when the canonical or anticanonical bundle is ample.
Surjectivity of is not necessary for the existence of a projective lift. In forthcoming work, we give more easily verified criteria for the unobstructedness of polarized deformations of primitive symplectic varieties.
1.4. An obstructed symplectic fourfold
The crystalline torsion-freeness assumption in Corollary C cannot be omitted, even when the Hodge–de Rham spectral sequence degenerates in every degree and a projective symplectic -lift exists.
Theorem E.
Let and . There exists a smooth projective symplectic fourfold over whose Hodge–de Rham spectral sequence degenerates at and for which
Both and are obstructed, although admits a smooth projective symplectic lift over .
Since trivializes , this gives a characteristic-three symplectic answer to the existence question raised in [8, Remark 1.1]: a variety with trivial canonical bundle and -degenerate Hodge–de Rham spectral sequence whose mixed-characteristic deformations are obstructed. Thus the crystalline torsion-freeness assumption in their Theorem A cannot be omitted without replacement. The same question is addressed by Zock using a bielliptic surface in characteristic two [34].
Proposition 8.5 constructs the fourfold as a free cyclic quotient of a product of ordinary elliptic curves and realizes three linearly independent geometric obstruction directions by a coherent Bockstein. In contrast, Proposition 8.6 gives an unobstructed smooth projective symplectic sixfold with nonzero degree-three crystalline torsion. Neither example is irreducible symplectic.
1.5. Outline of the proof
For a closed -form , let be the space-valued formal moduli problem of deformations of carrying a closed lift of . The period problem parametrizes lifts of the point determined by to marked maps ; its definition (3.7) requires no chosen integral period or rank-one direct summand.
Crystalline invariance gives a natural period map on animated divided-power bases. The Kaledin–Verbitsky complex computes the tangent object of (Lemma 4.2). Using divided-power envelopes of the diagonal, we identify the derivative with contraction by on vector fields and the filtered inclusion on forms. When
| (1.3) |
which happens precisely for volume and symplectic forms (Remark 6.3), the derivative is simply . The calculation on shifted square-zero tests also identifies the map on complete obstruction spaces.
The equality makes the relevant filtered maps strict, so injectivity of the joint contraction (1.1) detects the complete obstruction of a tuple. On the target, universal coefficients show that degree- period classes lift whenever is torsion-free. Brantner–Taelman’s bootstrapping theorem [8, Proposition 7.8] then extends the conclusion from divided-power elementary extensions to all Artinian extensions. At this step requires an initial -point: Achinger–Suh’s description of the conjugate differential [1, Theorem 1.3 and Corollary 4.4] supplies a geometric -lift, and a filtered Bockstein argument lifts the forms to it using only the hypotheses of the theorem. Throughout, we rely on the foundations for formal moduli problems and animated crystalline cohomology developed in [8, 5, 23].
1.6. Torsion and the scope of the criterion
The torsion hypotheses ensure that every period lifts across Artinian extensions, and they are necessary for this: by the explicit hull of Proposition 3.9, the classical truncation of the deformation problem of periods is unobstructed if and only if the adjacent crystalline cohomology group is torsion-free. This necessity concerns only the deformation problem of periods; it does not imply necessity for geometric unobstructedness. In the symplectic case, every actual obstruction lies in a canonical selective subspace, whose vanishing gives a refined criterion that tolerates some degree-three torsion (Section 8).
1.7. Relation with earlier work
Bogomolov proved unobstructedness for compact Kähler holomorphic symplectic manifolds [6], and Tian and Todorov did so for compact Kähler manifolds with trivial canonical bundle [30, 31]. Iacono–Manetti gave an algebraic proof for smooth projective varieties over an algebraically closed field of characteristic zero [16, Theorem 6.2 and Corollary 6.5]. Using the Cartan homotopy and Hodge–de Rham degeneration, their Theorem 6.4 places obstruction classes in the kernel of contraction against the cohomology of the canonical bundle; a volume form makes this test injective, and they show moreover that the governing differential graded Lie algebra is homotopy abelian. Theorem A keeps only the injectivity of contraction on from this mechanism, and it concludes unobstructedness and relative lifting of the forms rather than homotopy abelianity.
Kaledin–Verbitsky classify deformations of a smooth complex algebraic manifold with a closed nondegenerate two-form by their periods, which lie in the formal completion of at the class of the form, under an admissibility condition on every infinitesimal deformation [19, Definitions 2.1 and 2.4, Theorem 2.6]. The same truncated complex appears in our symplectic tangent calculation as . Properness allows us to impose all conditions on the special fibre: for strictness and degree-three torsion-freeness for lifting periods. We obtain smoothness; a classification by periods would also require control of the whole tangent fibre, including infinitesimal automorphisms.
Brantner–Taelman’s period map attaches to a deformation the arrow , whose fibre is the first step of the Hodge filtration [8, Definition 6.13]. Their Theorem 7.16 sharpens [8, Theorem A]: it assumes a splitting lifting the quotient , torsion-freeness of , and explicit conditions detecting obstructions, and thereby isolates the integral linear problem more precisely than torsion-freeness in all degrees [8, Remark 7.17]. At it requires a -lift, which their Theorem 7.18 supplies under the hypotheses of their Theorem A. Our deformation problem of periods (3.7) tracks a specified map , so its defining complex is and the integrality condition lies in the adjacent degree . We compute the derivative of our period map, which differs from the first-filtration derivative of [8, Theorem 6.18], in Propositions 4.4 and 4.6, and we construct the initial lift of the tuple at in Section 5.4.
Organization
Section 2 recalls formal deformation theory; Sections 3 and 4 construct the period map and compute its derivative, and Section 5 proves Theorem A. Sections 6 and 7 give applications and algebraization results, and Section 8 treats the selective refinement and torsion examples.
Conventions
We write and . We abbreviate , put and , and write for a coherent sheaf ; the argument is omitted when clear. Deformations in this paper are marked, that is, equipped with an identification of the special fibre with , and they are unpolarized unless a line bundle is specified. A deformation functor is unobstructed if it takes every surjection of such algebras to a surjection of sets.
We write for the set-valued classical deformation functor and for the space-valued geometric deformation problem, retaining automorphisms and higher homotopies. The deformation problems of pairs and of periods are also space-valued. The superscript denotes classical truncation, so ; the superscript used below refers instead to closed differential forms.
We use sans-serif letters for auxiliary -categories, calligraphic letters for sheaves, complexes, and algebraic lifts, and Fraktur letters for formal schemes. We write for cohomology and for sheaf hypercohomology; crystalline and filtered de Rham complexes are denoted by , , and .
Acknowledgement
The author thanks Lie Fu, Zhiyuan Li and Teppei Takamatsu for helpful discussions and comments.
AI declaration
This paper grew out of several research notes, written with the help of large language models (LLMs), aimed at understanding the formal deformation theory of Calabi–Yau varieties developed in [8]. We asked ChatGPT Pro to survey the classical algebraic proofs of the Bogomolov–Tian–Todorov theorem, which drew our attention to the deformation theory of symplectic varieties in characteristic zero due to Kaledin and Verbitsky [19]. It also supplied elementary deformation-theoretic computations suggesting that the argument should extend to positive and mixed characteristic under the natural assumptions of Hodge–de Rham degeneration and torsion-free crystalline cohomology. We then tried to construct animated period maps following the ideas of Kaledin and Verbitsky, but for arbitrary closed forms rather than only symplectic ones. Finally, we arrived at the divided-power Cartan homotopy formula of Proposition 4.4 with the help of ChatGPT 6 Astra and ChatGPT 6.1 Sol. ChatGPT was also used to investigate the case where is not torsion-free, and it produced the symplectic fourfold and sixfold of Section 8.3.
We also pushed ChatGPT 6 Astra to find useful refinements in the presence of torsion via prismatic deformation theory, but it did not produce interesting results beyond the classical framework.
The author drafted the paper. Claude Code was used to help with proofreading, readability, organization, and notation.
2. Preliminaries on formal deformation theory
We collect the conventions and results from [8, §§2–4, 6.1, 7.4] that we use to construct the deformation problems of varieties, pairs, and periods. Throughout, is a perfect field of characteristic and . Our tangent object is the tangent fibre of [8, Remark 3.11], with the same shift convention.
2.1. Categorical conventions
Write for the -category of spaces. For an -category , we denote its mapping spaces by and its core, consisting of objects and equivalences, by . All limits and colimits are taken in the relevant -category, so a point of includes a path between the images of its two components in . In particular, a point of the fibre includes an identification with ; in a deformation problem, this identification is the marking. Natural transformations include the compatible homotopies in their naturality diagrams [21, §§1.2, 4.2].
For an animated algebra , write for its stable -category of modules and for the perfect modules, namely the retracts of finite cell modules. For discrete , the homotopy category of is the unbounded derived category of -modules. We use cohomological shifts and hence . In particular, a discrete module is concentrated in cohomological degree . The underlying infinite loop space is denoted , so
| (2.1) |
In a stable -category, . Tensor products in stable module -categories are derived, and the symbol emphasizes this convention. Unadorned tensor products of discrete modules denote ordinary tensor products. Tensoring with a fixed module preserves finite limits and colimits, and preserves limits [22, §1.1]. The stable -category of decreasing filtered modules is , with and limits and colimits computed at each filtration step.
2.2. Animated algebras and cotangent complexes
The -category of animated commutative -algebras is obtained from simplicial commutative -algebras by inverting weak equivalences, or from polynomial -algebras by animation [8, §2.2]. In positive characteristic this -category differs from that of connective --algebras, but its module -categories are still stable.
Following [8, Definition 3.1], let be the -category of augmented animated -algebras such that is local Artinian with residue field and has finite length over . Write for the ordinary category of its discrete objects; for , let denote its maximal ideal. For a finite-dimensional -vector space and , the split square-zero algebra has augmentation ideal and -structure through , which is denoted by in [8]. The case , gives the discrete ring of dual numbers .
For in animated algebras, the cotangent complex represents derived derivations: for a connective -module ,
| (2.2) |
Here the maps on the left are sections of . Cotangent complexes satisfy transitivity and commute with derived base change [8, Corollary 2.18 and Propositions 2.19–2.20]. For a smooth map of discrete algebras , in degree zero. The same derivation formula holds for sheaves of animated algebras [8, Proposition 2.31].
2.3. Formal moduli problems and tangent objects
Definition 2.1.
A functor is a formal moduli problem if is contractible and the natural map
is an equivalence whenever is surjective on on both sides. We denote the resulting -category by . Its classical truncation is for .
This is [8, Definition 3.6]; the pullback condition is called infinitesimal cohesiveness, and it also holds when only one of the two maps is surjective on . By [8, Lemma 3.7], is a deformation functor: it sends classical pullback squares with one map surjective to surjections of sets, and to bijections when the common base is and the other ring is . The condition fixes the special object and its marking, but it does not imply that is nonempty for every .
The tangent object is characterized by natural equivalences of pointed spaces
| (2.3) |
for finite-dimensional and [8, Remark 3.11]. The basepoints come from the split maps , and the equivalences are compatible with the deloopings
More generally,
| (2.4) |
Thus records infinitesimal automorphisms, is the ordinary tangent space, and is the complete obstruction space described below.
A natural transformation induces a map . It is an equivalence of formal moduli problems if and only if it induces equivalences on all shifted square-zero tests [8, Proposition 3.9]. The criterion applies to a given transformation: an equivalence of bare tangent complexes does not construct one. The additional nonlinear operations are encoded by the partition Lie formalism [8, Remark 3.11]; see also [7]. Our arguments use the functorial tangent and obstruction constructions rather than this formalism.
2.4. Elementary extensions and complete obstructions
An elementary extension in is one whose kernel satisfies , so that is a finite-dimensional -vector space. By [8, Lemma 3.3], it is classified by a map in a homotopy pullback
| (2.5) |
The lower map is the zero section. Equivalently, the extension class is represented by an -linear map , where acts on through . The -structure on retains mixed-characteristic extension classes: , and the nonsplit extension gives a nonzero element of .
For and , applying to (2.5) identifies the marked lift space with
| (2.6) |
The obstruction class of is
| (2.7) |
A lift of on exists exactly when vanishes. Once a marked lift is chosen, the lift space is identified with ; its components form a torsor under , and its relative automorphisms are [8, §3.3].
These constructions are natural in . In particular, a transformation carries to by the induced map on , and gives a compatible map of marked lift spaces. Every surjection in factors into elementary extensions [8, Corollary 3.4]. Hence is unobstructed exactly when all its obstruction classes vanish [8, Proposition 3.14], and is a sufficient condition for this.
2.5. Divided-power bases and tangent tests
Give its canonical divided powers on . Animated divided-power rings are generated from divided-power polynomial rings by sifted colimits [8, Definition 4.4]. Let be the -category of animated divided-power -algebras whose divided-power reduction is and whose underlying augmented algebra belongs to [8, Definition 6.1]. A divided-power formal moduli problem satisfies Definition 2.1 with pullbacks taken in this -category [8, Definition 6.2]. The forgetful functor preserves limits, so is a divided-power formal moduli problem for every [8, Corollary 4.7 and Lemma 6.3].
For a connective perfect -module , the divided-power square-zero algebra has underlying animated algebra [8, Definition 4.32 and §6.3]. Consequently has the same tangent object as , with the same shifted tests and degree conventions. Restriction to divided-power bases should not be confused with deforming sheaves of divided-power rings; the two problems are related by the induced divided-power construction of [8, Construction 6.8].
A discrete divided-power extension with kernel is elementary when [8, Definition 7.7]
| (2.8) |
By [8, Proposition 7.15], its classifying square (2.5) exists in with target . Thus the same complete obstruction and lift-space descriptions apply to the restricted problems , using their divided-power square-zero tests. At , the extension is not elementary in this sense, because . This is why the bootstrapping theorem Theorem 5.5 has a separate initial-lift hypothesis.
2.6. Geometric deformations
For a -scheme , fix its small Zariski site and let be the -category of Zariski sheaves of animated algebras on . Let be the sheafification of the constant presheaf with value , and write . Consider the geometric deformation problem constructed as in [8, Construction 3.22]. The marked-fibre description of Remark 3.23 therein will serve as our working definition.
Lemma 2.2.
For with augmentation , the geometric deformation problem admits a natural equivalence
| (2.9) |
Thus its points are pairs , where and is a specified equivalence in , or equivalently pushout squares
A path from to consists of an equivalence in together with a specified homotopy ; all higher compatibility homotopies are retained. For an augmented map , base change sends this pair to , where
Proof.
In [8, Construction 3.22], consider
where the two maps to are and source evaluation, and is the first projection. Its objects are structure maps . For , the pushout universal property identifies -cocartesian transport with . An arrow over is therefore cocartesian exactly when the comparison is an equivalence, or equivalently its square in is a pushout. Thus the -subcategory of all objects and cocartesian arrows is the left fibration classified by ; this is the dual of [21, Corollary 2.4.2.5].
The fibre of over consists of cocartesian arrows , hence of equivalences . This gives (2.9) and the pushout description, with all pushouts computed in sheaves. Paths in this homotopy fibre are equivalences together with the stated marking homotopy and all higher compatibilities. For fixed , the space of markings is , which need not be contractible. Associativity of sheaf pushouts gives the displayed marking and naturality in . ∎
Proposition 2.3 ([8, Proposition 3.25]).
The geometric deformation problem is a formal moduli problem.
For a quasi-compact and quasi-separated scheme , the sheaf-algebra construction agrees with deformation in derived schemes, and its objects are flat over [8, Corollaries 3.32 and 3.35]. The flatness criterion states that a connective -module is flat if and only if its derived special fibre is discrete [8, Lemma 3.34]. For smooth proper and discrete , the resulting space is the usual groupoid of smooth proper marked deformations. Its classical truncation is therefore .
Let denote the stable -category of sheaves of -modules. We record the tangent calculation of [8, Corollary 3.26 and Remark 3.27], including the passage from square-zero tests to the tangent object.
Proposition 2.4.
For a smooth proper -scheme , there is a natural equivalence
| (2.10) |
Here is the derived internal sheaf Hom. Since , this specializes to
| (2.11) |
Consequently, the infinitesimal automorphism, tangent, and complete obstruction spaces are respectively , , and ; all three are finite-dimensional.
Proof.
Put . For a finite-dimensional -vector space and , [8, Corollary 3.26] and the internal-Hom adjunction give
Here is finite locally free, and the coefficient complex is a finite direct sum of shifts of , so its tensor product commutes with derived global sections. Coefficient naturality and compatibility with square-zero deloopings identify with by (2.3), as in [8, Remark 3.21]. Since , this proves both formulas; taking cohomology and using properness gives the remaining assertions. ∎
2.7. Descent
Let be the Čech nerve of a finite affine Zariski cover. Descent for sheaves of animated algebras and their equivalences gives
| (2.12) |
Here is the limit of the cosimplicial diagram in , and the local functors use the same marked sheaf-algebra construction. The limit includes overlap equivalences and all their compatibility homotopies. Mapping spaces and homotopy fibres preserve this descent, so the same procedure applies to the closed-form problems in Section 3.
Applying the tangent construction to all shifted square-zero tests linearizes this limit. On a smooth affine , the local tangent object is , and its descent limit is . More generally, limits in a stable -category preserve fibre sequences and hence cofibre sequences, so local tangent triangles and natural transformations globalize by taking the same descent limit. We use this procedure for the pair tangent object and the derivative of the period map in Section 4.
3. Deformations of closed forms and crystalline periods
Fix a smooth proper -scheme of dimension , an integer , and a closed form . We construct the deformation problem of marked pairs carrying closed lifts of , the period problem and its classical hull, and the crystalline period map on divided-power bases. The constructions require only smooth properness and closedness; further hypotheses are stated in the lifting results.
3.1. Relative de Rham complexes and closed forms
For and , let denote the Hodge-filtered relative derived de Rham complex, with derived -completion understood. We use Illusie’s construction [17, VIII, §2.1] in the polynomial-resolution form of [5, Definition 2.1 and Remark 2.2]: locally, we realize the de Rham complexes of a simplicial polynomial resolution, filtered by form degree . We write
Derived -completion is taken at each filtration step [5, Definition 8.1] and is automatic here because is nilpotent in . Hodge completion is a different operation [5, Remark 2.6]; it actually has no effect by the finite filtration proved below.
For the animated divided-power formulation and its compatibility with derived de Rham cohomology for animated rings, see [23, Definition 4.31 and Lemma 4.32]. On a divided-power base , the induced structure on gives the same filtered complex [8, Corollary 5.12].
Lemma 3.1.
The filtration on is finite, with and for . Its graded pieces, for , are
| (3.1) |
Here denotes the derived exterior power of the cotangent complex. Every is a perfect -module, and for every in , with the derived base change, there is a natural equivalence
| (3.2) |
For discrete , the filtered complex is with its Hodge filtration .
Proof.
The associated graded formula for the Hodge filtration is part of Illusie’s construction; see [2, §1.2] and, for the animated formulation, [8, Propositions 5.10 and 5.15]. The deformation is smooth and proper in the derived sense, and is finite locally free of rank at most . The polynomial-resolution construction commutes with derived base change, including each Hodge step; see [5, Proposition 2.7] and [23, Corollary 4.38]. Together with the smooth comparison modulo powers of [5, Corollary 3.10], this identifies, on each affine open, the reduction of with for every . Tensoring with is conservative on all -modules: the finite Postnikov tower of and composition series of its homotopy groups build from finitely many shifts and extensions of as an -module. Thus implies . Applying this to each , , proves finiteness of the filtration before taking global sections.
The exterior powers of are perfect, and their derived global sections are perfect and commute with derived base change by proper smooth pushforward. Applying this to (3.1) and using the finite filtration proves perfectness and (3.2). The identification for discrete is again [5, Corollary 3.10]; its applicability uses the nilpotence of on . ∎
Definition 3.2.
The space of closed relative -forms of degree zero on is
| (3.3) | ||||
The degree-zero argument is suppressed in the notation. This space is functorial in , including changes of animated base. Projection to gives its underlying-form map
On animated bases, a point retains the homotopies expressing closedness, rather than only an underlying differential form.
Lemma 3.3.
If is discrete, then is discrete and its points are precisely
Proof.
The complex has no cohomology below degree , and its degree- cohomology is the displayed kernel; apply (2.1). ∎
3.2. The deformation problem of the pair
For , its marking gives a reduction map . Define as the total space over whose fibre at is
| (3.4) |
Thus a point consists of a marked deformation , a closed form , and a path identifying its reduction with . Marked equivalences of transport the forms and their reduction paths. The total space is formed by the Grothendieck construction [21, §2.2], which defines and its forgetful map to functorially.
Proposition 3.4.
The functor is a formal moduli problem satisfying Zariski homotopy descent. On discrete bases it is the groupoid of marked pairs , and
| (3.5) |
Its tangent object is perfect, and forgetting the form gives a natural equivalence
| (3.6) |
In particular, is a deformation functor with finite-dimensional tangent space.
Proof.
The value at is contractible. Consider a pullback as in Definition 2.1. By the formal moduli property of , compatible geometric deformations glue to . The coefficient fibre sequence
remains a fibre sequence after tensoring with . By (3.2), the tensor product is identified with the corresponding filtered de Rham complex of . Applying the shift , then , and taking the fibre at proves the formal moduli axiom for (3.4), and hence for the total spaces. Sheaf descent for the filtered complexes, combined with (2.12), gives the same assertion on a Zariski cover.
Remark 3.5.
Once the geometric deformation has been glued, gluing its forms uses exactness in the stable module -category, as in the proof above; this argument does not require individual relative Hodge cohomology groups to be flat. If (1.3) holds at the special fibre, derived Nakayama shows that every infinitesimal lift also satisfies it.
3.3. The deformation problem of periods
Recall that crystalline cohomology is perfect and its derived reduction is de Rham cohomology:
The form defines a point by the composite . For any , set
and define the deformation problem of periods by
| (3.7) |
A point consists of a period and a path identifying its reduction with . This is defined on all animated Artinian bases.
Proposition 3.6.
The functor is a formal moduli problem, with
| (3.8) |
For finite-dimensional and , the split lift of gives a natural equivalence
Its ordinary tangent and complete obstruction spaces are and , respectively.
Proof.
The underlying -modules of an infinitesimal pullback of animated algebras form a homotopy pullback in : surjectivity on makes the homotopy pullback connective. Tensoring with , shifting by , applying , and taking the fibre at preserve this limit. Since is contractible, the functor is thus a formal moduli problem.
On a split test, the augmentation is split with fibre . Translation by the split lift of gives the stated equivalence, and hence (3.8). ∎
The reduction path remains part of a point of even for discrete , and this period space can have higher homotopies. It therefore records more than a set-theoretic fibre of degree- cohomology classes.
3.4. Hulls for the deformation problem of periods
Let and . We set
| (3.9) |
for any .
For any , the same argument as in Proposition 3.6 shows that the space-valued fibre defines a formal moduli problem; hence its classical truncation is a deformation functor. When and , this recovers from (3.7). Unobstructedness of means that is surjective for every surjection in .
Lemma 3.7.
For a perfect -complex and a discrete -module , there is a natural exact sequence
| (3.10) |
If is torsion-free, then naturally, and every surjection of discrete -modules induces a surjection .
Proof.
Since the DVR has global dimension one, the hyper-Tor spectral sequence has only the columns ; its degree- edge sequence is (3.10), also given by [29, Tag 0H80]. If is torsion-free, it is finite free, so the Tor term vanishes. Right exactness of tensor product then gives the asserted surjectivity, without any freeness assumption on . ∎
Lemma 3.8.
Let in . A point lifts to if and only if its underlying class in lifts to .
Proof.
Necessity follows by forgetting the marking. Conversely, a class lift gives a period over and a path for a representative of . The path marks at , and identifies its image with . ∎
For a complete Noetherian local -algebra with residue field , write . A hull is a formally smooth natural transformation inducing an isomorphism on tangent spaces [29, Tag 06IY].
A minimal representative of is a bounded complex of finite free -modules, quasi-isomorphic to , such that for every ; equivalently, has zero differential. Such a representative exists by [29, Tag 0BCC]: the complex constructed there has rank in degree , so it is bounded because is perfect. Concretely, starting with a bounded finite free representative, one successively cancels a contractible summand whenever a differential has a unit entry; see [29, Tag 00MT]. Each cancellation lowers the total rank, so the process terminates with all differential entries in .
Proposition 3.9.
Choose a minimal bounded finite free representative and Smith bases in which the nonzero entries of are , with . Then
Transport to this model and choose a lift of its vector in . Number the first coordinates by the nonzero Smith entries. A hull of is
| (3.11) |
The hull map sends satisfying these equations to the cycle with its constant reduction path to .
Proof.
Minimality gives zero differentials on , and hence the formula for . We first explain why the Smith form of splits off two-term summands of the entire complex. For a nonzero diagonal entry , the identity forces the corresponding row of to vanish, while forces the corresponding column of to vanish. Here cancellation of is valid because the terms are free over the domain . Thus each block in degrees is a direct summand of and contributes to . In the complementary complex the differential in degree is zero, so its degree- cohomology is a submodule of a free module and is torsion-free. This proves the asserted description of .
For a discrete , the marked classes have the description
| (3.12) |
Indeed, a reduction path is represented in degree over . Lifting it to and changing the cycle by its boundary makes the marking constant. Since has zero differential, higher homotopies do not change the vector representing this path. Changes between constant markings therefore have coefficients in , giving precisely the displayed denominator. Writing , the cycle equations are . Thus represents the numerator of (3.12).
For formal smoothness, let , fix such a cycle over , and prescribe a class in reducing to its image. Choose a representative of this class. Its reduction differs from by for some . Lift to . Then represents the same marked class and lifts exactly. On every differential of vanishes, so the hull map induces the tangent isomorphism . ∎
Torsion in arises from incoming summands in degrees . Each such summand contributes a free hull parameter whose values in are identified modulo , and these boundary identifications can make the hull map noninjective.
Corollary 3.10.
The deformation functor is unobstructed if and only if is torsion-free.
Proof.
A deformation functor admitting a hull is unobstructed exactly when that hull is formally smooth over , equivalently a power-series ring over [29, Tag 0DZK]. If , (3.11) is such a ring. If , choose . The element is nonzero, since its image in is nonzero, but it is killed by , which is impossible in a power-series ring over . Thus unobstructedness is equivalent to , which is precisely the asserted torsion-freeness. ∎
Corollary 3.11.
The classical truncation of the deformation problem of periods is unobstructed on all discrete Artinian -algebras if and only if is torsion-free.
Proof.
Apply Corollary 3.10 with and . ∎
Remark 3.12.
Under the torsion-free hypothesis of Corollary 3.11, every actual obstruction to lifting a period across a discrete elementary extension vanishes. The complete obstruction space of Proposition 3.6 can nevertheless be nonzero, since classical unobstructedness does not assert that is concentrated in degree zero. The necessity of torsion-freeness for the deformation problem of periods does not imply its necessity for unobstructedness of the geometric deformation functor.
3.5. Crystalline comparison and the period map
Give its canonical divided powers on . For divided-power bases, the crystalline property identifies the total relative complex with the fixed coefficient complex .
Lemma 3.13.
For and , there are natural equivalences of unfiltered module objects
| (3.13) |
They are compatible with marked equivalences of deformations, divided-power base change, and Zariski descent.
Proof.
Since the induced divided-power algebra of has divided-power reduction , [8, Corollary 5.12 and Theorem 5.22] give the first equivalence. For the second, form the pushout in sheaves of animated divided-power algebras. Its divided-power reduction is [8, Proposition 4.11]. Filtered base change and unfiltered crystalline invariance [8, Proposition 5.11 and Theorem 5.22] therefore identify with . Global-sections base change [8, Lemma 6.12] and crystalline comparison [8, Proposition 5.26] give . All these constructions are functorial, which gives the stated compatibilities. ∎
Remark 3.14.
Proposition 3.15.
Crystalline comparison defines a natural transformation of divided-power formal moduli problems
| (3.14) |
with the following properties:
-
(i)
It sends a marked pair over to
(3.15) with its induced reduction path to .
-
(ii)
It commutes with base change and Zariski homotopy descent; on discrete bases, its underlying cohomology class is the crystalline class of .
-
(iii)
For a finite-dimensional -vector space and , its restriction to the marked form fibre of the constant deformation over is
(3.16) induced by the Hodge inclusion.
Proof.
- (i)
-
(ii)
Naturality gives compatibility with divided-power base change, and passage to the Zariski descent limit gives compatibility with Zariski homotopy descent. The smooth comparison identifies the underlying cohomology class with the usual crystalline class.
-
(iii)
Take the constant geometric deformation and translate the form and period fibres by the constant lift of . Filtered base change identifies the resulting map with (3.16), induced by the Hodge inclusion.
∎
Formula (3.16) describes only the variation of the form on a constant deformation; Section 4 computes the full derivative, including the contribution from geometric deformations of underlying varieties.
Remark 3.16.
In general, (3.14) does not extend to a morphism on all animated Artinian bases. Indeed, take an elliptic curve, , and a nonzero invariant differential. For a discrete Artinian -algebra , translations of the constant pair by would map to the period automorphism group , where is its image under the proposed extension. Naturality would therefore give a homomorphism to the additive formal group. Compatibility on dual numbers with trivial higher divided powers and Theorem 4.6 force to be an isomorphism, given up to sign by contraction with . Thus is invertible as a power series, whereas
would force , contradicting finiteness of .
The same obstruction occurs for the symplectic surface , with and , where is a nonzero invariant differential on the elliptic curve and is the projection to . Translations in the first factor would give a homomorphism from to the additive formal group associated with , whose differential is, up to sign,
The class is nonzero by de Rham Künneth and the Hodge inclusion for . Composing with a -linear functional nonzero on this class yields the same contradiction.
4. The derivative of the period map
Keep the notation of Section 3. We compute the derivative of from the tangent extension of the pair and the crystalline comparison homotopy.
4.1. The tangent extension and the Kaledin–Verbitsky complex
Work in the stable -category of sheaves of -modules on . Put
Since is closed, is a -linear map of sheaf complexes. Here is concentrated in degree zero. Write for its cofibre. It is the middle row of the commutative diagram of complexes
whose columns are indexed by cohomological degree; here denotes the unshifted de Rham differential. The vertical maps are the canonical maps of the cofibre sequence of , and each column is split exact. As in [19, §6], we also use the complex
with in degree zero, in degree one, and the displayed positive differentials. We call it the Kaledin–Verbitsky complex of the closed form , or KV complex for short.
Remark 4.1.
Lemma 4.2.
There is a natural cofibre sequence in :
| (4.1) |
and an identification
| (4.2) |
Proof.
By Proposition 3.4 and (2.11), the fibre and quotient of the tangent extension are and . To identify the connecting map, trivialize the geometric deformation on smooth affine charts. Over the dual numbers , an infinitesimal automorphism changes the constant form by
Thus the infinitesimal action is . The deformation problem of a local pair is the homotopy quotient of the form fibre by changes of this trivialization. Linearizing this quotient gives the cofibre of the action map . Rotating the cofibre sequence of gives the connecting arrow in (4.1).
This action calculation is natural in the coefficient module. For a connective perfect module , choose a simplicial vector space representing it. On a polynomial chart , filtered base change identifies the relative complex in degree with . The universal derivation gives the same coefficient-linear Lie derivative in every simplicial degree. Realizing these module maps computes the action with coefficients in , including all shifted tests. Taking the stable descent limit of Section 2.7 proves (4.1). Changing the signs of the terms, with the convention made explicit in Remark 4.1, identifies with and gives (4.2). ∎
4.2. The crystalline comparison homotopy
By Proposition 3.15, the derivative on the form fibre is induced by the filtered inclusion
We use for the inclusion of the indicated Hodge step into the total de Rham complex. For every in the stable -category , the universal property of the cofibre gives a pullback square of spaces
| (4.3) |
Here is precomposition with , and the bottom arrow selects the zero morphism. Extending to the pair tangent object therefore requires a choice of nullhomotopy of . For the period map, this homotopy comes from Lemma 3.13. Cartan’s formula gives
The following calculation identifies contraction with the homotopy induced by that crystalline–de Rham comparison, including its naturality on shifted divided-power tests.
Lemma 4.3.
Let , where is a discrete -vector space with trivial higher divided powers, and put
The vertex maps and induce . They admit a homotopy satisfying
Here sets . For every -derivation , evaluation at sends to the contraction , a homotopy from the identity to .
Proof.
Write and put
with in degrees zero and one. Since , the decomposition of differentials into the and gives an identification of complexes
Here the differential on a tensor , with , is . Thus [29, Tag 07LC] applies in the -direction with coefficient ring and coefficient module in each -degree . We now make its contraction and the signs of the total differential explicit.
Let set , let include the constants, and put . The identity for shows that the -linear degree-minus-one map
satisfies and . On use the tensor homotopy
with the Koszul sign convention for tensor products of graded maps. The identities for the telescope to . Moreover, . Extend this to the total complex by
This sign cancels the two terms involving in . Since includes the constants in and augments , we obtain and . Finally , so satisfies and .
Put and , and equip with the PD ideal generated by and the . Evaluation , is a map of PD de Rham complexes: it is induced by a PD map , since the higher divided powers on vanish. Give weight zero and assign weights to , respectively. Both and preserve weight. After evaluation, every term of weight at least two vanishes: , the coefficients of lie in , and for . The weight-zero part is killed by , as are the weight-one terms containing a but no . For a weight-one term containing in the th differential position, moving it past the remaining differentials and applying gives the sign and replaces by . Evaluation then gives . Thus the evaluated homotopy is contraction:
| (4.4) |
Evaluation sends to and to the identity, so
| (4.5) |
∎
Locally on , let for a finite-dimensional -vector space , and write for the constant lift of . For an automorphism of the constant deformation over reducing to the identity, the reduction-compatible naturality of (3.13) gives a path from the identity to on the total de Rham complex. Evaluating this path at and translating by gives a path from to . Linearizing these paths in such automorphisms gives a path
The following proposition identifies this homotopy with contraction.
Proposition 4.4.
The nullhomotopy of in constructed above is given, before shifting, by contraction
| (4.6) |
with . In the shifted target, this nullhomotopy is represented by the graded map of degree , given by
Here the sign comes from the differential on . This identification is compatible with restriction and natural on for every connective perfect -module .
Proof.
Work first on a polynomial chart, with as in Lemma 4.3. Filtered base change and the induced divided-power comparison [8, Proposition 5.11 and Corollary 5.12] identify the polynomial and PD-polynomial complexes in that lemma with
Put and . The natural unfiltered equivalences of [8, Theorem 5.22] give
with and , including their natural coherences. These are the reduction equivalences used in Lemma 3.13; [8, Proposition 5.26] identifies their globalization with Berthelot’s crystalline cohomology. The classical computation by PD diagonal envelopes is given in [29, Tag 07LG and Tag 07LH].
Here is an equivalence, either by the explicit contraction of Lemma 4.3 or by the displayed reduction equivalences. Consequently
has contractible homotopy fibre over . The vertices belong to this fibre with their constant augmentation paths, and says that is a path between them in this fibre. Postcomposition with identifies it with the contractible fibre
The comparison path is the path in this latter fibre obtained from the natural reduction maps: both vertices reduce to , and their paths to are those supplied by the same augmentation diagram. Thus its specified augmentation normalization agrees with that of , and contractibility identifies the two paths together with their coherences. For an -derivation , put and let fix and send to . It satisfies with these same reduction coherences. It therefore takes this identification to the comparison path from to on , whose homotopy is by (4.4). Here the substitution and composes the universal homotopy with the map of de Rham complexes defined by , giving . For a closed -form on , let be its pullback to . Applying this comparison homotopy to means precomposing its path with . Its chain representative is ; closedness gives .
This uniqueness also holds for parameter diagrams. Fix the polynomial algebra , and let index a diagram of discrete coefficient spaces and derivations , with coefficient maps carrying each derivation to the next. All the preceding arrows, their reductions, and their evaluations form diagrams of -modules, and the formula for is natural in these coefficient maps. Since and are objectwise equivalences, they are equivalences in . The same mapping-fibre argument in this -category, with the diagram as source, identifies with the comparison path as a path of natural transformations. In particular, it supplies all compatibilities in the coefficient and derivation parameters.
For a connective perfect -module , choose a simplicial vector space representing it, with each finite-dimensional. The full derived derivation space on is
Since is finite free, this space is computed by the simplicial spaces of degreewise derivations into . Apply the preceding argument to these parameter diagrams, with , , and . The functors and preserve geometric realizations [8, Definition 4.32, Corollary 4.7, and Definition 5.6]. Realizing the diagrams of divided-power arrows and their comparison maps therefore gives the corresponding diagram over , together with its augmentation normalization. In every simplicial degree,
The evaluated homotopy is linear in the coefficients , so its realization is contraction with coefficients in . The diagramwise identification above makes it the comparison homotopy for every such , including every shifted test.
Finally, cover by smooth affine opens admitting étale coordinates. The evaluated contraction is intrinsic and commutes with localization and étale pullback, so these formulas agree on coordinate refinements. Since the crystalline comparison and its normalized identification are natural on the same refinements, they define the same morphism of sheaf tangent complexes, which descends to by Section 2.7. Evaluation at the closed form gives and (4.6). ∎
Remark 4.5.
The animated divided-power comparison supplies the higher coherences, and the polynomial calculation identifies its derivative. The passage from divided-power elementary extensions to arbitrary Artinian extensions uses the bootstrapping argument of Section 5.
4.3. The tangent morphism
Combining the Hodge inclusion with Proposition 4.4 gives the full derivative.
Theorem 4.6.
With the sign convention of Remark 4.1, the tangent morphism of the period map has the following descriptions.
- (i)
-
(ii)
If contraction is an isomorphism, as in (1.3), this morphism identifies with the filtered inclusion
(4.8) For every finite-dimensional -vector space and , the period map on is then
Proof.
By Lemma 4.2, the source tangent object is . The derivative restricts to on the form fibre by (3.16), and Proposition 4.4 identifies its comparison nullhomotopy with . The cofibre property (4.3) therefore gives a chain map
with component in degree and the identity on in every degree . Transporting along and shifting by gives , with the vertical maps shown in the diagram. Naturality in connective perfect coefficients and descent, proved in Proposition 4.4, identify with the tangent morphism of the period map on every shifted divided-power test.
If contraction is an isomorphism, the same components define a chain isomorphism with . Taking derived global sections gives (4.8). Under this identification, the tangent triangle (4.1) is the filtered triangle
Here the geometric quotient is identified with the last term by . Finally, tensoring the tangent morphism with and applying gives the formula on divided-power square-zero tests. ∎
4.4. Tangent and obstruction spaces
With our conventions, degrees , , and of a tangent object describe infinitesimal automorphisms, tangent vectors, and complete obstructions, respectively.
Corollary 4.7.
The period map induces the maps
Both are induced by . If (1.3) holds, the infinitesimal automorphism, tangent, and complete obstruction spaces of the pair are, respectively,
| (4.9) | ||||
and their maps to the corresponding spaces of are induced by the filtered inclusion .
Proof.
Restriction to divided-power bases preserves the tangent objects, because has underlying animated algebra . Taking and of the morphism in Theorem 4.6 therefore gives the two displayed maps. Under (1.3), taking , , and in (4.8) gives (4.9) and the maps induced by the filtered inclusion. ∎
For a divided-power elementary extension with kernel satisfying (2.8), the classifying square of Section 2 places each pair obstruction in . The case therefore identifies the map on actual obstruction classes with that of Corollary 4.7 tensored with ; its crystalline boundary description is Proposition 5.1. Complete obstruction spaces may remain nonzero when every actual class vanishes (Remark 3.12).
5. Proof of the main criterion
The proof of Theorem A combines injectivity on complete obstruction spaces, unobstructedness of periods, and Brantner–Taelman bootstrapping. At , this requires the initial tuple lift constructed in Section 5.4.
5.1. Naturality of complete obstructions
Let be a divided-power elementary extension with kernel , and let . By Section 2, both restrictions in (3.14) send its classifying square to homotopy pullbacks, and naturality gives a homotopy-commutative diagram
| (5.1) |
Its lower map is the derivative on the shifted square-zero test with coefficient module . On connected components, it is the complete obstruction map of Corollary 4.7 tensored with .
On the target, the coefficient extension supplies the distinguished triangle
| (5.2) |
After shifting by and applying the mapping-space construction, its connecting map is the obstruction map for the deformation problem of periods. We denote the induced boundary map on cohomology by
Proposition 5.1.
Under (1.3), the complete pair obstruction satisfies
| (5.3) |
Here is the underlying crystalline period class, and the pair obstruction belongs to .
Proof.
Take connected components of (5.1), use Corollary 4.7 for the lower map, and use (5.2) for the target obstruction. Completeness of the source obstruction follows from its lift-space description (2.6), since is a formal moduli problem. ∎
Remark 5.2.
Identity (5.3) is induced by a morphism of lift-space homotopy pullbacks, so it is compatible with automorphisms and changes of representatives.
Remark 5.3.
Under (1.3), classical Čech–de Rham descent also gives a complete obstruction in ; we neither use this construction nor identify its class with .
5.2. Degreewise Hodge conditions
The following lemma translates the numerical equality into the filtered injectivity and surjectivity used below.
Lemma 5.4.
Let be a bounded complex over a field with a finite filtration and finite-dimensional cohomology in every graded piece. Suppose
For every , the maps are injective, and both
are surjective.
Proof.
Choose a bounded finite-dimensional, degreewise split filtered model whose graded complexes have zero differential. Such a model is obtained by representing the successive filtration extensions by cofibres of chain maps and cancelling the contractible pairs in their diagonal graded blocks; these cancellations preserve the filtration. Thus
For a finite-dimensional complex,
The hypothesis therefore forces . In degree , the asserted maps are inclusions and quotients of the split filtered vector space . In degree , every representative of a class in lifts to an element of , which is a cycle since . This proves the remaining surjectivity. ∎
5.3. From divided-power bases to all Artinian bases
We use the following bootstrapping theorem of Brantner–Taelman [8, Proposition 7.8].
Theorem 5.5.
Assume that is infinite. Let be a deformation functor with finite-dimensional tangent space. Suppose is surjective for every divided-power elementary extension. If , suppose also that . Then is unobstructed.
Its mixed-characteristic truncated divided-power test rings detect equations in a hull, whereas testing only divided-power -algebras would not suffice.
5.4. The initial lift in characteristic two
To supply the initial tuple point required by Theorem 5.5 at , we first lift forms on a fixed -lift by a filtered Bockstein argument, and then construct such a lift using the conjugate spectral sequence. The first step works for every .
Lemma 5.6.
Let be a smooth proper lift of . If holds and is torsion-free, then every is closed and has a closed relative lift on .
Proof.
Put and , with their stupid filtrations. Smooth proper filtered base change gives . Tensoring the coefficient sequence over with therefore gives a distinguished triangle
The filtered inclusions give a morphism from this triangle to the corresponding triangle for , and hence a natural square of Bocksteins
Crystalline–de Rham comparison and derived crystalline base change identify for , where , compatibly with reduction [29, Tag 07MW, Tag 07MU]. These identifications are unfiltered: under them, the total coefficient triangle is , whereas the filtered triangle uses the filtration of the specified lift and the filtered base change above. By universal coefficients and torsion-freeness of , the map is surjective, so . By Lemma 5.4 in degree , the map is injective. The same numerical equality kills , so is closed and belongs to . Naturality yields , hence . The filtered coefficient sequence lifts to , which is the kernel of . ∎
Proposition 5.7.
Suppose (1.1) is injective and holds for every . Then admits a smooth proper lift to .
Proof.
Put and write for the base change of . The conjugate spectral sequence is
Since is perfect, the sum of the dimensions of the terms of the conjugate spectral sequence in total degree equals the corresponding Hodge dimension sum, which by already equals the dimension of the abutment. Passing to the next page decreases a total-degree dimension sum by the ranks of the incoming and outgoing differentials. Consequently, all differentials touching this diagonal vanish; in particular, .
Let . By [1, Theorem 1.3 and Corollary 4.4],
To identify the operation, note that [1, §2.2 and Corollary 4.3] describe the underlying extension-class map as exterior comultiplication followed by , whose evaluation on a decomposable form is the usual alternating sum of contractions. No factor or appears, so the identity also holds at .
Flat base change by the field automorphism preserves the injectivity of (1.1). Hence , and the complete obstruction theory of smooth schemes gives a -lift of . Base change by the inverse Witt Frobenius automorphism of yields a lift of , which is smooth and proper because these properties persist across the nilpotent thickening. ∎
Corollary 5.8.
Under the hypotheses of Proposition 5.7, assume also torsion-freeness of for every . Then the tuple has a lift to with all the forms closed.
Proof.
Choose one geometric lift from Proposition 5.7 and apply Lemma 5.6 to each on this same lift. The lifts of the individual forms are chosen independently, and together they lift all the prescribed forms simultaneously. ∎
The fixed-lift comparison used here is valid for the canonical divided powers on : the smooth-lift comparison of [29, Tag 07MW] requires nilpotent in the base, without requiring its divided powers to be nilpotent. At , the identity means that is not a divided-power elementary extension. This does not affect the comparison just used, and the constructed pair lift supplies the separate initial point required by Theorem 5.5.
5.5. Lifting forms on a prescribed deformation
Lemma 5.9.
Let be a morphism of space-valued formal moduli problems over . If is unobstructed and is surjective, then for every surjection of discrete Artinian bases the map
is surjective on connected components. In particular, is formally smooth on discrete bases, and is unobstructed; no hull assumption is needed.
Proof.
This is the space-valued analogue of the classical criterion [29, Tag 0E3R]. It suffices to consider an elementary extension with kernel . Fix in the displayed homotopy fibre product, where identifies with . By (2.6), the components of the marked lift spaces over and are functorial torsors under and , respectively. The first is nonempty by unobstructedness, and tangent surjectivity therefore supplies a lift mapping to compatibly with . Factorization into elementary extensions proves formal smoothness. Taking shows that every -point has a preimage in ; lifting such a preimage proves unobstructedness of . ∎
5.6. Joint detection of obstructions
Proof of Theorem A.
First, every is closed: the Hodge differential enters total degree , so forces it to vanish.
The tuple deformation problem is the finite homotopy fibre product of the individual closed-form problems over . It is a formal moduli problem by closure under finite limits, and its tangent cohomology is finite-dimensional by properness and the bounded coherent terms below. Put
and let be their respective direct sums. If denotes the tuple tangent object, Lemmas 4.2 and 4.6 give a morphism of triangles
Indeed, the action of a vector field on each closed form is , so its action morphism factors through the connecting morphism of the corresponding filtered triangle. Composing the middle vertical arrow with the sum of the filtered inclusions gives the derivative of the product period map.
By Lemma 5.4, applied in degree , both connecting maps and vanish. The triangles therefore yield a diagram of short exact sequences
The right map is joint contraction followed by the diagonal signs from Proposition 4.4; it is injective by assumption, so the middle map is injective. Moreover, since is injective by Lemma 5.4, the tuple period map is injective on complete obstruction spaces.
By universal coefficients and the torsion-freeness of the indicated crystalline groups, the product of the deformation problems of periods is classically unobstructed. Naturality and completeness of obstruction classes then imply that the tuple functor is unobstructed across divided-power elementary extensions. At , Corollary 5.8 supplies the initial tuple lift. Brantner–Taelman bootstrapping then gives unobstructedness across all Artinian surjections when is infinite.
For finite , extend scalars to and . Proper base change and flatness preserve all the hypotheses, and when the same construction supplies the initial point over ; hence the result holds over the infinite field . For an elementary extension over with kernel , the complete obstruction lies in , its formation commutes with scalar extension, and its image over vanishes. Since extension of scalars is injective on this finite-dimensional vector space, the original obstruction vanishes, and factorization into elementary extensions gives unobstructedness over .
Finally, the boundary factors through the zero map , so is surjective. Applied to the morphism from the tuple problem to , Lemma 5.9 gives formal smoothness on discrete bases and unobstructedness of . ∎
6. Volume forms, symplectic forms, and examples
We deduce Corollaries B and C from Theorem A and then give examples of the joint criterion, including varieties with a nontrivial canonical bundle and a nonzero geometric obstruction space. This section and the next depend on Sections 2, 3, 4 and 5 only through the statement of Theorem A.
6.1. Varieties with trivial canonical bundle
Let be a smooth proper variety of dimension over .
Lemma 6.1.
Integral crystalline Poincaré duality for gives perfect torsion pairings (with Frobenius twists omitted):
| (6.1) |
In particular, is torsion-free if and only if is torsion-free.
Proof.
Integral crystalline duality gives
Apply Berthelot’s crystalline Poincaré duality [3, VII, §2.1] with coefficient crystal over each . The finite-level trace pairings are perfect, and passing to the compatible perfect complexes over gives the displayed equivalence; equivalently, the trace-induced map is an equivalence modulo , and hence over by derived Nakayama. No freeness of individual crystalline cohomology groups is assumed here. Since has global dimension one, the universal-coefficient sequence for the derived dual is
The last term is free, and for a finite torsion module , . Thus the torsion in the middle is exactly the dual of the torsion in degree . This proves the claim; the extra in this degree is essential. ∎
Proof of Corollary B.
6.2. Symplectic varieties
Lemma 6.2.
If is an isomorphism for a two-form , then and . Under , .
Proof.
The corresponding nondegenerate alternating pairing has even rank and admits local symplectic bases. Changes between these bases have determinant one; equivalently, the integral Pfaffian polynomial transforms as a top exterior form. The resulting nowhere-vanishing section trivializes . This argument also works at and when divides , since the relation is not used to define .
Finally, the map is the differential of the Hodge–de Rham spectral sequence. Since it enters total degree three, forces it to vanish. ∎
Proof of Corollary C.
The contraction isomorphism is injective on , so Theorem A applies with . The assumed numerical equality makes closed, and degree-three crystalline torsion-freeness is the only integral hypothesis required, even at . Nondegeneracy persists on every lift by Nakayama’s lemma. ∎
Remark 6.3.
Let be a closed -form with . Invertibility of contraction in (1.3) forces , hence or when , since for . For , the identification makes this the nowhere-vanishing volume-form condition. For , it makes symplectic, so is even and is trivial by Lemma 6.2. In dimension one only the volume case occurs; in dimension two the cases coincide, and always forces . The joint injectivity on in Theorem A imposes no such bundle-rank restriction, permitting the examples with nontrivial canonical bundle in Section 6.3.
Example 6.4.
Let be an abelian variety of dimension . A basis of invariant one-forms identifies with by joint contraction. The Hodge–de Rham spectral sequence degenerates [25, Proposition 5.1], and crystalline cohomology is the exterior algebra on the free module [18, II, §7.1]. Thus Theorem A applies in every characteristic. It recovers unobstructedness of the unpolarized deformation functor, whose tangent space has dimension , and also gives relative lifting of the chosen invariant basis. With these one-forms, the criterion uses crystalline degree two in every characteristic, independently of .
Example 6.5.
For a K3 surface , crystalline cohomology is free, the Hodge–de Rham spectral sequence degenerates, and , while ; see [11, Proposition 1.1]. A generator of is a nondegenerate two-form. Hence Corollary C recovers unobstructedness over , including in characteristic two, and relative lifting of this form. Here the usual geometric obstruction space already vanishes; the products below give examples in which it does not. If is ample and , Serre duality makes surjective, so Theorem D also gives unobstructed deformations of the polarized surface with its two-form and a smooth projective -lift. Projectivity of a specified formal lift requires an additional condition; see Theorem 7.4.
Moduli spaces of stable sheaves give higher-dimensional examples. For the following proposition and examples, assume that is algebraically closed. If is a K3 or abelian surface, write for the moduli space of Gieseker -stable sheaves with Mukai vector , where is the rank and in the K3 case, while in the abelian case. We put and call -general if every -semistable sheaf of vector is stable. On an abelian surface the determinant is allowed to vary within its numerical class.
Proposition 6.6.
Let , let be a K3 surface, and let be primitive with and . Suppose that is -general and that admits a smooth projective lift over . Then is a smooth projective symplectic variety of dimension , and
Consequently, Corollary C applies to and its Mukai two-form, with no bound on .
Proof.
The geometric assertions and the relative moduli construction are given in [12, Proposition 4.4 and its proof]. The chosen lift of gives a smooth projective moduli space with special fibre ; generality of the polarization persists on the geometric generic fibre. After passage to characteristic zero, this fibre is of K3[n] deformation type, where . The Betti-number formula and integral torsion-freeness for Hilbert schemes of K3 surfaces [32, Theorems 1.1 and 3.1] therefore give
The adjacent degree-four assertion ensures that . Fontaine–Messing comparison in degree , in the form recalled in [24, Theorem 0.3], now gives . The crystalline universal-coefficient sequence
and Nakayama’s lemma give the vanishing of crystalline cohomology. Finally, the lift modulo and the truncated Deligne–Illusie decomposition [10, Theorem 2.1] imply
since . Thus holds, and Corollary C applies. ∎
The unramified lifting hypothesis in Proposition 6.6 is part of the criterion: the existence of a lift over a ramified extension of alone does not justify this comparison argument. The following examples reach the smaller primes by direct cohomological calculations.
Example 6.7 (K3 Hilbert squares in odd characteristic).
Let be a K3 surface and . Its Hilbert square is the stable-sheaf moduli space
for any polarization , and carries the Mukai two-form [12, Proposition 4.4]. Let be the blowup of along the diagonal. There is a finite degree-two morphism . The blowup formula adds the cohomology of shifted by to that of ; it holds integrally for crystalline cohomology and for Hodge cohomology [13, IV, Theorem 1.1.9, Corollary 1.1.11, and Theorem 1.2.1]. Together with Künneth and Example 6.5, it gives
Pullback followed by pushforward is multiplication by two [13, II, Proposition 4.2.9 and its proof], so pullback is injective in these theories. The same vanishing therefore holds on . Its degree-three de Rham cohomology vanishes as well, so holds. Thus Corollary C applies, including in characteristic three, without a lifting assumption on the polarized surface.
Example 6.8 (Sheaves on abelian surfaces at small primes).
Let be an abelian surface and its dual abelian variety. First suppose that , that is primitive with and , and that is -general. Then is a smooth projective sixfold with a Mukai two-form. Its Albanese fibre is a K3 surface, and base change of the Albanese map by on gives a finite étale cover
see [12, Proposition 6.9(i) and diagram (50) in its proof]. To see that is a K3 surface, note that the smooth proper lift constructed there has a K3 geometric generic fibre; hence is connected and . Its symplectic form trivializes , so . By Examples 6.4 and 6.5 and Künneth, the covering variety has torsion-free crystalline cohomology and -degeneration. Since is invertible in , normalized trace splits pullback on crystalline cohomology and on the filtered de Rham complex. Both properties descend to , proving the hypotheses of Corollary C, also when . For instance, gives .
There is also a four-dimensional stable-sheaf example in characteristic two. In every characteristic,
Indeed, a rank-one torsion-free sheaf with these invariants has double dual and quotient of length one, and is stable for every polarization. The sum of nonzero invariant two-forms from the two abelian surfaces is nondegenerate, including at . The cohomological assertions follow from Example 6.4, so Corollary C applies here as well. These examples verify the theorem’s hypotheses at the stated primes; they do not assert them for arbitrary sheaf moduli spaces in characteristics two and three.
6.3. Joint contraction on products
For forms on , let denote the joint contraction map from to the sum of the .
Lemma 6.9.
Let be smooth proper geometrically connected varieties, each equipped with a finite tuple of forms. Suppose that is injective for and every . Then the pullbacks of all these forms to jointly detect .
If every has -degeneration and torsion-free crystalline cohomology in all degrees, then has both properties as well. In this case Theorem A applies to the pulled-back tuple in every characteristic.
Proof.
Write . The tangent bundle splits by factors, and coherent Künneth gives
Forms pulled back from contract to zero on the other tangent factors, and Künneth identifies their joint contraction on the displayed summands with . These maps are injective; for one-forms, the separately indexed targets for the forms keep the tangent factors distinct.
The crystalline Künneth theorem and freeness give the integral assertion. Coherent and de Rham Künneth give -degeneration on the product by comparing dimensions in each total degree. ∎
Products of smooth proper geometrically connected curves of genus at least two have factorwise deformation theory over Artinian -algebras [4, Theorem 3.3], although with at least two factors [4, Warning 3.5]. The same splitting holds over by induction on elementary extensions: lifts the factors, while and Künneth identify with and split the lift torsors accordingly. Relative canonical sections are free and commute with base change, so the chosen one-forms lift to every prescribed deformation and their pullbacks are closed. These conclusions require no nonhyperellipticity assumption; the following proposition determines when joint detection holds.
Proposition 6.10.
Let be smooth proper geometrically connected curves of genera . Assume that each has genus two or is geometrically nonhyperelliptic. On , take the pullbacks of bases of . These forms satisfy the hypotheses of Theorem A in every characteristic. Moreover, is ample and
| (6.2) | ||||
| (6.3) |
Proof.
The canonical multiplication map
is surjective by Max Noether’s theorem, including genus two in every characteristic [33, Theorem 3.2.2]. By Serre duality, surjectivity is precisely the injectivity of
In degrees zero and two, the source vanishes.
Smooth proper curves have -degeneration [25, Corollary 5.2]. Their crystalline cohomology is torsion-free: the universal-coefficient sequence in degree zero, together with and , first gives freeness of , and Lemma 6.1 then gives freeness of . Apply Lemma 6.9. The dimension formulas follow from Künneth and . The canonical bundle is ample, being the external tensor product of the ample bundles . ∎
Example 6.11.
For two smooth plane quartics , the six pulled-back one-forms detect the obstruction space of , where
Since , canonical multiplication is simply multiplication of linear forms into quadrics on . No single one-form can detect the -dimensional space through contraction into the -dimensional . Thus Theorem A recovers classical unobstructedness by joint detection on a variety with ample canonical bundle.
For comparison, let be hyperelliptic of genus and let satisfy Proposition 6.10, with genus . Canonical multiplication on has cokernel of dimension [33, §3.3]. Serre duality and Künneth therefore give a kernel of dimension for joint contraction by the pulled-back canonical bases on . The variety and its tuple of forms remain unobstructed: the detection criterion is sufficient but not necessary.
Example 6.12.
Let be a K3 surface, and let satisfy Proposition 6.10. On , take the two-form from and bases of one-forms from and . Contraction with the K3 form is an isomorphism of tangent and cotangent bundles, so Lemma 6.9 applies, and both the tuple and the underlying variety are unobstructed over in every characteristic.
The K3 form detects the summands from , and the curve forms those from and ; neither subcollection detects all of . This illustrates joint detection by forms of different degrees on a variety whose nontrivial canonical bundle is pulled back from the ample canonical bundle of .
7. Polarizations and projective lifts
7.1. Lifting a polarization and algebraization
Proposition 7.1.
Assume the hypotheses of Theorem A. Suppose also that is projective and , and let be an ample line bundle on . Then deformations of are unobstructed, and forgetting is formally smooth on discrete Artinian bases. Moreover, every compatible formal lift of over is the completion of a smooth projective -scheme carrying closed relative lifts of the forms and a lift of .
Proof.
Let be an elementary extension with kernel , and fix a lift of a deformation . The exact sequence of units for this thickening gives an obstruction to lifting a line bundle on in
see [29, Tag 0C6R]. Since ampleness persists across a nilpotent thickening, factorization into elementary extensions proves the relative formal smoothness, and the unobstructedness assertion then follows from Theorem A.
Along a compatible system of -lifts, lift recursively with restriction identifications. This gives an invertible sheaf, ample on the special fibre, on the resulting proper adic formal scheme. Grothendieck’s algebraization theorem [14, Theorem 5.4.5] supplies a projective scheme with this completion and line bundle. It is smooth near the special fibre, and its closed nonsmooth locus is proper over , so it would meet the special fibre if nonempty; hence is smooth.
Coherent formal GAGA algebraizes each compatible section of . Their exterior derivatives vanish modulo every and hence vanish, since is a finite, -adically separated -module. ∎
In particular, a projective variety of dimension at least three with trivial canonical bundle and has a smooth projective -lift whenever it satisfies the hypotheses of Corollary B.
7.2. Choosing a lift that preserves a polarization
We replace the vanishing assumption in Proposition 7.1 by surjectivity of , allowing the geometric lift to vary; prescribed formal lifts are treated separately below.
Proof of Theorem D.
Let be an elementary extension with kernel , and let deform over . Choose a marked smooth proper lift using Theorem A. The exact sequence of units gives a complete obstruction
to extending to ; see [29, Tag 0C6R]. Changing the marked geometric lift by changes this obstruction by
| (7.1) |
for compatible choices of signs.
To verify this also in mixed characteristic, choose affine trivializations of , with lifted geometric and line-bundle transition maps and . The obstruction is represented by . Replacing by for a Čech cocycle representing changes this discrepancy, under , by
Here all coefficients multiplying reduce to , since . This proves (7.1); compare also [26, §1.12 and Corollary 1.14].
Since is surjective, we may choose to cancel , and then lifts. If prescribed closed forms are also part of the data, the relative formal smoothness in Theorem A lifts them to this modified geometric lift. Factoring an arbitrary Artinian surjection into elementary extensions proves both unobstructedness assertions.
Recursively choose actual lifts and restriction identifications over . The resulting proper formal scheme carries a compatible invertible sheaf restricting to . The algebraization and formal GAGA argument in Proposition 7.1 supplies the asserted smooth projective scheme, ample line bundle, and closed forms. ∎
Remark 7.2.
The proof chooses the geometric lift at each step, and it does not algebraize every prescribed formal lift. Indeed, the tangent image of is , so this morphism is not formally smooth when .
7.3. A prescribed formal lift
Let be a smooth proper adic formal scheme over with special fibre , and put . Write
Here the differential forms are continuous, and the filtration is the stupid filtration of their de Rham complex. Crystalline comparison gives an isomorphism
The filtration step depends on the chosen lift.
Proposition 7.3.
For , the following are equivalent:
-
(i)
Some extends to an invertible sheaf on .
-
(ii)
belongs to .
More precisely, let annihilate the torsion of the finite -module . If (ii) holds, then extends for , and extends for .
Proof.
This is the formal -adic Lefschetz criterion; we give the obstruction argument, including the issue in characteristic two. The compatibility of the logarithmic boundary with crystalline first Chern classes is established in [20, Lemma 3.2.7]; the same reference, Theorem 4.2.1, proves the assertion for flat proper formal schemes of ramification index . Here .
For , the exact sequence
and the convergent logarithm give an exact obstruction sequence
| (7.2) |
After inverting , the image of in equals the de Rham projection of . This compatibility can be seen in the Čech construction: logarithms of the triple-overlap discrepancies of lifted transition functions are the degree-zero components of the crystalline Chern cocycle. The exact sequence of complexes shows that the kernel of this projection is precisely ; no Hodge–de Rham degeneration is required.
Consequently, (ii) says that vanishes after inverting . Since is flat, multiplication by identifies with . Thus the torsion of the target in (7.2) is annihilated by , so , and exactness proves the claimed extension.
For , the obstruction to extending from to lies in and is killed by . Hence has an extension on . Now use
Functoriality of crystalline Chern classes and their comparison maps identifies the rational image of this boundary with the projection of ; this is the comparison for the intermediate thickening in the proof of [20, Theorem 4.2.1]. Under (ii), the boundary of therefore vanishes after inverting . Since the torsion of the target is killed by , the sheaf extends to , and its restriction to is .
Conversely, the de Rham Chern class of an invertible sheaf on lies in and restricts to its crystalline Chern class. Dividing by proves (ii) from (i). Finally, the finiteness of formal coherent cohomology used above is [14, Theorem 3.4.2]. ∎
Theorem 7.4.
Suppose the special fibre is projective. Then is the completion of a smooth projective -scheme if and only if there is an ample such that
| (7.3) |
In that case, every compatible closed relative form on algebraizes. In particular, every such with is projectively algebraizable.
Proof.
Under (7.3), Proposition 7.3 supplies an invertible sheaf whose special-fibre restriction is a positive tensor power of , hence ample. Apply [14, Theorem 5.4.5]. Smoothness and the algebraization of closed forms follow exactly as in the proof of Proposition 7.1 (no assumption on is required). Conversely, restrict a relatively ample line bundle on a projective algebraization and use the necessity in Proposition 7.3. If the rational second coherent cohomology vanishes, de Rham projection is zero, so its kernel contains every degree-two crystalline class after comparison. ∎
Corollary 7.5.
Let be smooth and proper. If either or is ample, every smooth proper formal lift of has a smooth projective algebraization. Compatible closed relative forms on algebraize as well.
Proof.
The relative canonical bundle or its inverse restricts to an ample line bundle on . Grothendieck’s algebraization theorem [14, Theorem 5.4.5] gives a projective algebraization; smoothness and the algebraization of closed forms follow as in Proposition 7.1. ∎
For example, every prescribed smooth proper formal lift of a product of curves in Proposition 6.10 is projectively algebraizable. With two or more factors, is nonzero, so this does not follow from Proposition 7.1; instead, the canonical line bundle lifts to every deformation, whether or not is surjective.
8. Crystalline torsion: a refinement and two examples
The torsion-freeness hypothesis makes the deformation problem of periods classically unobstructed. For unobstructedness of geometric deformations this is stronger than necessary, because the derivative sees only a filtered part of de Rham cohomology. We describe the intersection that can contain actual obstructions, and in Section 8.3 we construct two symplectic varieties in characteristic three with torsion in but opposite deformation behaviour. Throughout this section, is a smooth proper symplectic pair over , and
We assume that is closed; this follows automatically from whenever that condition is imposed below.
8.1. The support of coefficient boundaries
Reduction and universal coefficients give a canonical injection
| (8.1) |
Indeed, is free, so . Let denote the image of (8.1). Thus , and is defined without choosing a decomposition of the crystalline complex.
Lemma 8.1.
For every elementary extension in , with kernel , the coefficient boundary satisfies
Proof.
Represent by a bounded complex of finite free -modules. Successive Smith decompositions split it into free one-term summands and two-term summands . Only a two-term summand in degrees two and three can contribute to the displayed boundary. Its degree-two coefficient cohomology is , and its connecting map is
where lifts . The expression belongs to and is independent of the lift because . Its target is the reduction of the upper term of that torsion summand, so its image lies in . All other summands have zero boundary in this degree. Summing over the summands proves the assertion, and the support itself is the canonical subspace defined above. ∎
8.2. The selective obstruction space
Write
Here uses the Hodge filtration under the displayed de Rham identification, and is induced by projection to form degree zero. By Theorem 4.6, is the complete obstruction space of the pair problem. Set .
Proposition 8.2.
Every pair obstruction over a divided-power elementary extension with kernel belongs to . There is a canonical exact sequence
| (8.2) |
Under , identifies this space with , where .
Proof.
By Proposition 5.1, the image of the complete pair obstruction is the coefficient boundary of its period. Lemma 8.1 therefore places it in , proving the first assertion. The triangle identifies with and with . Taking the inverse image of gives (8.2). Under , Lemma 5.4 makes injective, so its restriction identifies with the stated intersection. ∎
Corollary 8.3.
Assume and
| (8.3) |
Then and are unobstructed over , and forgetting the form is formally smooth on discrete Artinian bases. This holds for every prime .
Proof.
By Proposition 8.2 and completeness, every obstruction across a divided-power elementary extension vanishes. At , we supply the initial pair separately. Since contraction by is an isomorphism, Proposition 5.7 gives a smooth proper geometric lift . The filtered coefficient sequence for its de Rham complex gives a Bockstein
By crystalline comparison and filtered naturality, the image of the Bockstein of in is the total coefficient boundary for . Lemma 8.1 places this image in , and the filtered construction places it in , so it vanishes by (8.3). Since is injective by Lemma 5.4, the filtered Bockstein vanishes. Hence lifts to a closed relative two-form on . Its contraction remains invertible by Nakayama’s lemma.
For infinite , Theorem 5.5 now proves unobstructedness of the pair functor. For finite , apply this argument over ; the complete obstruction spaces commute with scalar extension, so their actual classes vanish already over . Finally, Lemma 5.4 gives the surjection , so Lemma 5.9 proves both the relative assertion and unobstructedness of the underlying variety. ∎
Condition (8.3) concerns the position of the torsion reduction relative to the Hodge filtration rather than its length or exponent, and taking recovers the symplectic criterion of Corollary C. On the other hand, a nonzero is only a space of possible obstruction values, and its nonvanishing does not prove that an actual obstruction is realized. When this space is nonzero, the argument asserts neither the continuation of every finite pair nor the existence of a compatible ramified lifting branch; these questions require additional control of the period and its coefficient boundaries.
8.3. Two projective examples with crystalline torsion
We exhibit both possible behaviours of the classical deformation functor when has torsion. The obstructed example admits a projective symplectic -lift, so the existence of one lift must be distinguished from the unobstructedness of the deformation functor. Throughout these examples, and . For a smooth proper -scheme , write
Universal coefficients give
Write for crystalline Frobenius, with its semilinearity over Witt-vector Frobenius understood, and for the cyclic group of order three, whose generator acts by the indicated automorphism in each construction.
We use the elliptic scheme
Its discriminant is . The tangent line meets the cubic with multiplicity three at , so has order three. Moreover, the points remain distinct modulo three. We also write for its base change to , and for its special fibre. The three distinct torsion points show that is ordinary. Put .
Ordinarity makes Cartier bijective on an invariant volume form, so Frobenius duality gives a normalized splitting of Frobenius on an ordinary abelian variety. Since its canonical bundle is trivial, the space of splitting maps is the one-dimensional space of sections of , and normalization fixes the scalar. The splitting is therefore invariant under every automorphism and descends through any finite étale quotient, including one of degree three. Petrov’s decomposition theorem then gives full Hodge–de Rham degeneration on these quotients [27, Theorem 1.1].
These quotients are also Bloch–Kato ordinary: for every , the sheaf of exact -forms has vanishing cohomology in all degrees [8, Definition 8.8]. Indeed, exact forms commute with pullback on the étale sites, so for either abelian cover the Cartan–Leray spectral sequence is
Every term on its second page vanishes because is ordinary. This proves ordinarity of the quotient also when the covering degree is three.
All quotients below arise from free actions on abelian schemes over . They are smooth and proper, and projectivity follows by tensoring an ample bundle over its finite group orbit and descending the resulting linearized ample bundle. For both crystalline cohomology and relative Hodge cohomology, pullback followed by trace is multiplication by three, and the corresponding groups on the abelian cover are -free. Hence three kills all torsion in the crystalline and relative Hodge groups of either quotient.
Lemma 8.4.
Let be a smooth proper scheme over a perfect field of characteristic . If crystalline Frobenius is bijective on , then
Proof.
The reduction of crystalline Frobenius is induced on the de Rham complex by in form degree zero and by zero in positive form degrees, with the corresponding Frobenius twist on the coefficient field understood. This is functorial derived base change; it does not replace derived reduction by reduction of an individual crystalline cohomology group. In particular, it kills the image of . On the canonical torsion reduction , it is the reduction of Frobenius on and is therefore bijective, because the universal-coefficient injection is Frobenius-equivariant, even when has torsion. A vector in the stated intersection is thus killed by this map and lies in a subspace on which the map is injective, so it is zero. ∎
8.4. An obstructed fourfold admitting a projective lift
Let act on by translation by on the first factor, by
on the second and third factors, and by the identity on the fourth. Since , this defines an order-three action. Each nonidentity element translates the first factor by a nonzero section, so the action is free over . Put
For an invariant differential on , let denote its pullback from the th factor. The form
is invariant, closed and nondegenerate; invariance on the middle factors follows from . It therefore descends to a relative symplectic form on , and we write for its reduction.
Proposition 8.5.
The smooth projective symplectic fourfold satisfies
Both and are obstructed over . More precisely, three linearly independent geometric obstruction directions are realized on an elementary extension with one-dimensional kernel. Nevertheless, the pair has the projective symplectic lift just constructed.
Proof.
We first compute the cohomology, without using the crystalline obstruction comparison. Put
Then , and is an abelian-surface fibration trivialized by the connected étale torsor . Modulo three, the matrix is a nontrivial unipotent block of size two, and its associated vector bundle is a nonsplit extension
Indeed, the extension cocycle is the image in of the nonzero torsor class in . The Artin–Schreier sequence makes this map injective, because is surjective. Thus . Descent gives
Since is a curve, there are no Leray differentials. Hence , and product Künneth gives
| (8.4) |
Write , with . The representation on invariant one-forms of the cover is . Taking global sections after descent gives
since . To compute the remaining Hodge number, we use
In characteristic three, . The regular block corresponds to the direct image of under , so . The projection formula and Leray now give , and Künneth with gives . Consequently
| (8.5) |
In the rational degree-one cohomology of the cover, the trivial character has multiplicity four and each nontrivial character of has multiplicity two. Rational descent and exterior powers therefore give and . Hodge degeneration gives de Rham dimensions and in degrees one and two. Since by universal coefficients in degree zero, the dimension formula gives and . Since the trace argument above shows that the torsion is killed by three, this proves the asserted modules.
To prove actual obstructedness, put . These modules also have torsion killed by three. After extending the generic field to contain a primitive cube root of unity, let be a nontrivial character of . On the generic fibre of the cover, both invariant one-forms and have characters . Therefore
Since is torsion-free, coherent universal coefficients, (8.5), and give
It follows that . Contraction with identifies this module with .
Let . The coefficient Bockstein
for factors as the universal-coefficient surjection onto , followed by its reduction into . Its image is therefore the canonical reduction of . This reduction is injective since the torsion has exponent three, so .
Choose a first-order deformation with tangent and glue it to the fixed lift over
The map
is an elementary extension with kernel . A lift over would extend to a square-zero deformation of , hence lift its class to . Its complete obstruction is exactly . Moreover, a reduction isomorphism over identifies the -restriction of any proposed lift with , so the necessity just proved applies to every lift in the absolute deformation functor, including lifts without a separately chosen -marking. Every vector in the three-dimensional image of is therefore realized as an actual geometric obstruction on this extension.
Finally, by contraction and the ordinary Cartan formula, the tangent object of the pair is identified with ; this identification uses only changes of local coordinates and forms. Hodge degeneration and Lemma 5.4 make its tangent map
surjective. Choose a pair tangent above and glue it to the fixed pair. If , this pair cannot lift over because its underlying variety cannot lift. This proves both obstructedness assertions using ordinary coherent deformation theory. ∎
Proposition 8.5 gives an affirmative answer in characteristic three to the existence question raised in [8, Remark 1.1]. Indeed, trivializes , the Hodge–de Rham spectral sequence degenerates at by the Frobenius-splitting argument above, and is obstructed. Thus Hodge–de Rham degeneration alone does not imply the conclusion of [8, Theorem A], even for a smooth projective symplectic variety admitting a projective symplectic -lift. For a related bielliptic example in characteristic two addressing the same question, see [34]. Here Calabi–Yau is understood in the sense of [8], namely triviality of the canonical bundle; the fourfold does not satisfy the additional vanishing for , as (8.4) shows.
8.5. An unobstructed sixfold with nonzero torsion
The second construction changes the integral representation on the cover while retaining its ordinary elliptic factors. Define
Let cyclically permute the three coordinates. It preserves the diagonal subgroup and hence acts on . Set
The norm is the class of , which is zero; thus . The sum map descends because , and it satisfies . Both nonidentity powers of therefore act freely. Put and . The abelian variety is ordinary because it is isogenous to .
Let , with and , be the invariant differentials on the two copies of which pull back to the coordinate differentials on . They exist because the isogeny is étale. The form
is invariant, closed and nondegenerate, and it descends to . We write for its reduction.
Proposition 8.6.
The smooth projective symplectic sixfold satisfies
Crystalline Frobenius is bijective on the latter module. Both and are unobstructed over , and forgetting the form is formally smooth on discrete Artinian bases.
Proof.
We give the Hodge and integral representation calculations needed for the conclusion. Let be the augmentation lattice and put . On étale Tate modules the isogeny adjoins to . Taking dual lattices gives, equivariantly,
The last decomposition follows by subtracting times . The sum map has the section , so for . Pullback from is the primitive summand , and therefore as a -lattice.
For an ordinary abelian variety, its unit-root crystalline summand is ; reduction of this summand maps isomorphically to . These identifications are compatible with endomorphisms [25, 18]. Since , we obtain
The map induced by becomes a product with fibre after the étale base change . The descent action can include translations, but these act trivially on the coherent cohomology of abelian varieties, so the degree-one fibre representation is . For a representation associated to this elliptic torsor, the descended bundle has and : the first equality follows by descent of global sections, and the second by Riemann–Roch for a degree-zero bundle. Using and , one finds
Coherent Leray therefore gives and .
The representation on one-forms of is . Its exterior square is . For the covering map , regular representation descent gives
Consequently
Rationally, is a sum of four regular representations. The invariant part in exterior degree two has dimension ; hence and . Hodge degeneration gives de Rham dimensions and in degrees one and two. As above, , and the dimension formula gives . Since the trace argument shows that the torsion is killed by three, this proves the asserted crystalline modules.
It remains to determine Frobenius on degree-three torsion. The ordinary connected–étale decomposition gives a -stable integral decomposition
Translations act trivially on cohomology, and the linear action preserves this canonical decomposition. The ordinary Hodge filtration identifies with . Lifting two regular generators of this last representation gives a map that is an isomorphism modulo three, so Nakayama and equality of ranks show that .
Crystalline étale descent along gives the Frobenius-compatible Cartan–Leray spectral sequence
One can equally construct it from the étale Čech nerve of the lifted torsor and crystalline–de Rham comparison. Frobenius compatibility comes from crystalline Frobenius on that nerve, so no Frobenius lift of is needed. The crystalline cohomology algebra of the abelian variety is the exterior algebra on . A diagonal tensor product of a regular module with any -free module is induced. Moreover, : unordered pairs in each regular basis form free orbits, and the cross tensor product is a sum of three regular modules. Thus , , and have no positive group cohomology.
It follows that the positive-column terms in total degree three are precisely
They have finite length and bijective Frobenius. Every Frobenius-stable submodule and quotient of a finite-length Frobenius-bijective module remains Frobenius-bijective: injectivity on a stable submodule and equality of lengths imply surjectivity there, and the quotient follows. Therefore their subquotients remain Frobenius-bijective. The positive-column filtration step of has finite length, while its quotient is a submodule of the free module , so this filtration step is exactly the torsion subgroup. Since bijectivity of Frobenius is preserved under extensions of finite-length modules, Frobenius is bijective on .
Now Lemma 8.4 gives . Hodge degeneration and Corollary 8.3 prove unobstructedness of both classical deformation functors and formal smoothness of forgetting the form. ∎
8.6. Comparison with Serre–Tate theory
Brantner–Taelman’s ordinary theory gives a further interpretation of these examples, beyond their crystalline splitting criterion [8, Theorem 7.16]. Their Serre–Tate theorem applies to a smooth proper geometrically irreducible Bloch–Kato -ordinary variety of dimension with trivial canonical bundle when is torsion-free [8, Theorem 8.28]. It describes the deformation problem by a formal group of multiplicative type, which can contain nonsmooth factors ; ordinarity alone therefore does not imply unobstructedness [8, Remark 1.5]. Over our algebraically closed field, their description of the classical deformation functor involves
with coefficient module [8, Remark 8.30]; when the latter is free of rank one, a choice of generator identifies with the character module. Here denotes de Rham–Witt Frobenius.
We first check the top-degree hypothesis for both examples: let be either or , and put . The invariant volume forms on the abelian covers descend, so and . The determinant of the action on the degree-one -adic étale cohomology of each cover is one, and rational descent consequently gives
Bloch–Kato -ordinarity gives
by [8, Proposition 8.11]. The universal-coefficient injection of into then shows that is free of rank one. Thus their Serre–Tate theorem applies to both examples; the fourfold’s obstructedness is compatible with the possible nonsmooth finite factors.
For the sixfold, we can also verify smoothness of the Serre–Tate group directly. Retain the notation , , and from the proof of Proposition 8.6. On the ordinary abelian cover, the integral Hodge–Witt decomposition and the exterior algebra description of crystalline cohomology give
compatibly with the -action and with [18, II, Remark 4.10(a), (4.10.1), and § 7.1, (7.1.1)]. In particular, is bijective on each of these modules. Since , every displayed module is induced and has vanishing positive group cohomology. Étale descent for de Rham–Witt cohomology therefore gives
This is a free -module of rank , with bijective . Taking -fixed elements over algebraically closed gives [18, II, Lemma 5.3]. By [8, Theorem 8.28 and Remark 8.30], the unpolarized deformation functor is consequently a formal torus, noncanonically isomorphic to , and is unobstructed.
Thus unpolarized unobstructedness of the sixfold is also a consequence of Brantner–Taelman’s full ordinary theory. The construction supplies explicit crystalline torsion and illustrates the selective criterion; our argument also proves relative lifting of the specified symplectic form, through formal smoothness of .
References
- [1] (2023) Some refinements of the Deligne–Illusie theorem. Algebra & Number Theory 17 (2), pp. 465–496. Note: arXiv:2003.09857v2 External Links: Document Cited by: §1.5, §5.4.
- [2] (2012) -adic periods and derived de Rham cohomology. Journal of the American Mathematical Society 25 (3), pp. 715–738. External Links: Document, Link Cited by: §3.1.
- [3] (1974) Cohomologie cristalline des schémas de caractéristique . Lecture Notes in Mathematics, Vol. 407, Springer-Verlag, Berlin. External Links: Document Cited by: §6.1.
- [4] (2013) Moduli of products of stable varieties. Compositio Mathematica 149, pp. 2036–2070. External Links: Document Cited by: §6.3.
- [5] (2012) -adic derived de Rham cohomology. Note: arXiv:1204.6560v1 External Links: Link Cited by: §1.5, §3.1, §3.1, §3.1.
- [6] (1978) Hamiltonian Kählerian manifolds. Doklady Akademii Nauk SSSR 243 (5), pp. 1101–1104. Note: In Russian External Links: Link Cited by: §1.7.
- [7] (2019) Deformation theory and partition Lie algebras. Note: arXiv:1904.07352 External Links: Link Cited by: §2.3.
- [8] (2025) Deformations and lifts of Calabi–Yau varieties in characteristic . Note: arXiv:2407.09256v3, 25 May 2025 External Links: Link Cited by: §1.4, §1.5, §1.7, §1, §1, §2.2, §2.2, §2.2, §2.3, §2.3, §2.3, §2.4, §2.4, §2.4, §2.5, §2.5, §2.5, §2.6, §2.6, §2.6, §2.6, §2.6, Proposition 2.3, §2, §3.1, §3.1, §3.2, §3.5, Remark 3.14, §4.2, §4.2, §5.3, §8.3, §8.4, §8.6, §8.6, §8.6.
- [9] (2009) Small resolutions and non-liftable Calabi–Yau threefolds. Manuscripta Mathematica 130 (2), pp. 233–249. External Links: Document Cited by: §1.
- [10] (1987) Relèvements modulo et décomposition du complexe de de Rham. Inventiones Mathematicae 89 (2), pp. 247–270. External Links: Document Cited by: §6.2.
- [11] (1981) Relèvement des surfaces K3 en caractéristique nulle. In Surfaces algébriques, Lecture Notes in Mathematics, Vol. 868, pp. 58–79. Note: Rédigé par Luc Illusie Cited by: Example 6.5.
- [12] (2021) Supersingular irreducible symplectic varieties. In Rationality of Varieties, Progress in Mathematics, Vol. 342, pp. 147–200. External Links: Document Cited by: §6.2, Example 6.7, Example 6.8.
- [13] (1985) Classes de Chern et classes de cycles en cohomologie de Hodge–Witt logarithmique. Mémoires de la Société Mathématique de France (N.S.) 21, pp. 1–87. External Links: Document Cited by: Example 6.7.
- [14] (1961) Éléments de géométrie algébrique. III. étude cohomologique des faisceaux cohérents. I. Publications Mathématiques de l’IHÉS 11, pp. 5–167. Note: Avec la collaboration de Jean Dieudonné Cited by: §7.1, §7.3, §7.3, §7.3.
- [15] (1999) A non-liftable Calabi–Yau threefold in characteristic . Tohoku Mathematical Journal (2) 51 (4), pp. 479–487. External Links: Document Cited by: §1.
- [16] (2010) An algebraic proof of Bogomolov–Tian–Todorov theorem. In Deformation Spaces: Perspectives on Algebro-Geometric Moduli, H. Abbaspour, M. Marcolli, and T. Tradler (Eds.), Aspects of Mathematics, pp. 113–133. Note: Theorem numbers refer to arXiv:0902.0732v2 External Links: Document Cited by: §1.7.
- [17] (1972) Complexe cotangent et déformations II. Lecture Notes in Mathematics, Vol. 283, Springer-Verlag, Berlin–New York. External Links: Document Cited by: §3.1.
- [18] (1979) Complexe de de Rham–Witt et cohomologie cristalline. Annales Scientifiques de l’École Normale Supérieure (4) 12, pp. 501–661. External Links: Link Cited by: Example 6.4, §8.5, §8.6.
- [19] (2002) Period map for non-compact holomorphically symplectic manifolds. Geometric and Functional Analysis 12 (6), pp. 1265–1295. External Links: Link Cited by: §1.7, §1, §4.1.
- [20] (2025) Lifting line bundles to schemes over -adic base. Note: Preprint, 18 November 2025 External Links: Link Cited by: §7.3, §7.3.
- [21] (2009) Higher topos theory. Annals of Mathematics Studies, Vol. 170, Princeton University Press. Note: Author version dated 9 April 2017 External Links: Link Cited by: §2.1, §2.6, §3.2.
- [22] (2017) Higher algebra. Note: Version dated 18 September 2017 External Links: Link Cited by: §2.1.
- [23] (2024) Revisiting derived crystalline cohomology. Bulletin de la Société Mathématique de France 152 (4), pp. 659–784. External Links: Document, Link Cited by: §1.5, §3.1, §3.1.
- [24] (2020) Integral -adic Hodge theory of formal schemes in low ramification. Note: Version 2, 9 December 2020 External Links: 2004.04436, Link Cited by: §6.2.
- [25] (1969) The first de Rham cohomology group and Dieudonné modules. Annales Scientifiques de l’École Normale Supérieure (4) 2, pp. 63–135. External Links: Link Cited by: §6.3, Example 6.4, §8.5.
- [26] (1979) Supersingular K3 crystals. Astérisque 64, pp. 3–86. External Links: Link Cited by: §7.2.
- [27] (2025) Decomposition of de Rham complex for quasi--split varieties. Note: Version 1, 19 February 2025 External Links: 2502.13356, Link Cited by: §8.3.
- [28] (2026) On the witt liftability of quasi-f-split calabi–yau varieties. External Links: 2610.38427, Link Cited by: footnote 1.
- [29] (2026) Stacks Project. External Links: Link Cited by: §3.4, §3.4, §3.4, §3.4, §4.2, §4.2, §5.4, §5.4, §5.5, §7.1, §7.2.
- [30] (1987) Smoothness of the universal deformation space of compact Calabi–Yau manifolds and its Peterson–Weil metric. In Mathematical Aspects of String Theory, Advanced Series in Mathematical Physics, Vol. 1, pp. 629–646. External Links: Document Cited by: §1.7.
- [31] (1989) The Weil–Petersson geometry of the moduli space of (Calabi–Yau) manifolds. I. Communications in Mathematical Physics 126 (2), pp. 325–346. External Links: Document Cited by: §1.7.
- [32] (2020) The integral cohomology of the Hilbert scheme of points on a surface. Forum of Mathematics, Sigma 8. External Links: Document Cited by: §6.2.
- [33] (2022) The canonical ring of a stacky curve. Memoirs of the American Mathematical Society 277 (1362). Note: Corrected author version, June 2023 External Links: Link Cited by: §6.3, Example 6.11.
- [34] A variety with trivial canonical bundle, degenerate Hodge-to-de Rham spectral sequence, and obstructed deformations. Note: Undated note External Links: Link Cited by: §1.4, §8.4.