Obstruction Theory for Hodge Classes

• Obstruction theory for Hodge classes is a framework defining criteria that prevent a Hodge cohomology class from extending or being realized by an algebraic cycle.
• It integrates diverse methods including intersection cohomology on Shimura varieties, logarithmic extensions in toroidal varieties, p-adic liftability, and local deformation analyses.
• These approaches reveal that standard Hodge conditions are insufficient, necessitating refined invariants like automorphic spectra, filtered Frobenius maps, and Bockstein obstructions to test algebraicity.

Searching arXiv for recent and relevant papers on obstruction theory for Hodge classes. Obstruction theory for Hodge classes studies criteria and mechanisms that prevent a cohomology class of Hodge type from arising, extending, lifting, or being realized by an algebraic cycle under a specified geometric or arithmetic construction. In recent work, the subject appears in several distinct but related settings: intersection cohomology on Shimura varieties, logarithmic Hodge theory on toroidal varieties, $p$-adic and crystalline approaches to liftability of Tate classes, local deformation theory for divisor classes, and integral or torsion obstructions detected by Bockstein constructions and perverse-sheaf methods. Across these settings, obstruction theory refines the Hodge problem from a question of existence to a question of realization, extension, and constructive access (Mostaed, 6 Feb 2026, Luo, 9 Sep 2025, Costa et al., 2020, Biswas et al., 2020, Rahman, 4 May 2026).

1. Conceptual scope and basic forms of obstruction

A Hodge class is typically a rational cohomology class lying in a $(p,p)$-summand, or in a corresponding filtered or integral analogue depending on the cohomology theory under consideration. Obstruction theory asks whether such a class can be extended across a boundary, lifted from characteristic $p$, deformed in a family, or represented by an algebraic cycle.

Several distinct obstruction formalisms now coexist. In logarithmic Hodge theory for toroidal varieties, the obstruction is a boundary morphism from Hodge classes on the open part to hypercohomology of an obstruction sheaf constructed from a mapping cone (Luo, 9 Sep 2025). In $p$-adic geometry, the obstruction is the failure of a crystalline Tate class to lie in the pulled-back Hodge filtration, expressed through a map of filtered $\phi$-modules (Costa et al., 2020). For divisors in first-order deformations, the classical obstruction in $H^2(X,\mathcal O_X)$ is refined to a local topological obstruction in $H^3(X,\mathcal O_X)$, or equivalently $H^2_D(X,\mathcal O_X)$, which carries strictly more information (Biswas et al., 2020). In the integral setting, Bockstein constructions produce torsion Hodge classes whose non-algebraicity is detected through an obstruction channel built from MacPherson–Vilonen gluing and Brauer data (Rahman, 4 May 2026).

A further variant arises in automorphic geometry. On orthogonal Shimura varieties, a rational Hodge class may be explicitly constructed in intersection cohomology while lying outside interior cohomology; this non-interiority obstructs its production by special cycles, theta lifts of cycles, endoscopic constructions, or boundary pushforwards, all of which yield interior classes (Mostaed, 6 Feb 2026). This suggests a broader taxonomy in which obstructions are not only cohomological but also spectral and representation-theoretic.

2. Automorphic obstruction on Shimura varieties

An explicit automorphic obstruction is developed for the Shimura variety attached to $G=\mathrm{SO}(V,Q)$ with $\dim_{\mathbb Q}V=28$ and signature $(p,p)$0. Let $(p,p)$1 denote the Baily–Borel compactification. The construction produces a rational class

$(p,p)$2

which lies in the middle Hodge summand

$(p,p)$3

Thus $(p,p)$4 is a rational Hodge class in intersection cohomology (Mostaed, 6 Feb 2026).

The class is obtained from a stable residual automorphic representation. The starting point is the unique weight-$(p,p)$5 newform of conductor $(p,p)$6,

$(p,p)$7

with Hecke field $(p,p)$8. Its attached adjoint Galois representation $(p,p)$9 is $p$0-dimensional, has Hodge–Tate weights $p$1, conductor $p$2, and is Steinberg at $p$3. The theta lift $p$4 to $p$5 is nonzero by Rallis’s inner-product formula, and its Arthur parameter is

$p$6

This parameter is residual and stable. By Vogan–Zuckerman, the archimedean component has nonzero $p$7-cohomology only in degree $p$8, and the infinitesimal character forces Hodge type $p$9 (Mostaed, 6 Feb 2026).

The decisive obstruction is non-interiority. Interior cohomology is defined by

$p$0

The main non-interiority statement is

$p$1

By Franke’s theorem, interior cohomology receives contributions only from cuspidal automorphic representations, whereas non-cuspidal spectrum contributes to full intersection cohomology but not to $p$2. Since $p$3 comes from a stable residual representation, it cannot be interior (Mostaed, 6 Feb 2026).

This non-interiority has direct geometric consequences. Any algebraic cycle $p$4 on $p$5 with $p$6 defines a class in $p$7. Moreover, special cycles, Hecke correspondences, theta lifts of cycles from smaller groups, endoscopic Shimura subvarieties, and boundary Gysin pushforwards all contribute through interior or non-stable channels. Since $p$8 is stable residual and non-interior, it cannot arise from any of these known constructions (Mostaed, 6 Feb 2026).

This does not disprove the Hodge conjecture. The explicit result is instead an obstruction to realization by currently known cycle constructions. A plausible implication is that automorphic cohomology can exhibit Hodge classes that are visible cohomologically but inaccessible through established geometric recipes.

3. Logarithmic and toroidal obstruction theory

For a smooth complex variety $p$9 with $\phi$0 a reduced SNC divisor and $\phi$1 toroidal, logarithmic Hodge theory provides a boundary-sensitive obstruction formalism. The logarithmic de Rham complex

$\phi$2

carries the Hodge filtration, and Deligne’s theorem together with Wei’s refinement gives $\phi$3-degeneration for the Hodge-to-de Rham spectral sequence associated to $\phi$4 (Luo, 9 Sep 2025).

The obstruction is defined from the mapping cone

$\phi$5

Its cohomology sheaves are supported on the boundary $\phi$6. The $\phi$7th obstruction sheaf is

$\phi$8

Filtering by $\phi$9 yields, for $H^2(X,\mathcal O_X)$0, an exact segment

$H^2(X,\mathcal O_X)$1

For $H^2(X,\mathcal O_X)$2, the primary obstruction is

$H^2(X,\mathcal O_X)$3

When $H^2(X,\mathcal O_X)$4, vanishing of $H^2(X,\mathcal O_X)$5 is equivalent to extension from a Hodge class on $H^2(X,\mathcal O_X)$6 (Luo, 9 Sep 2025).

The theory is developed further for rational-weighted toroidal varieties $H^2(X,\mathcal O_X)$7, where the weight function is $H^2(X,\mathcal O_X)$8-linear on maximal cones and satisfies a piecewise-linearity and convexity compatibility across faces. In this setting, local obstructions at points of $H^2(X,\mathcal O_X)$9 are analyzed via cone complexes $H^3(X,\mathcal O_X)$0, and compatibility of the rational weights forces the relevant obstruction evaluations to vanish in the weighted toroidal setting (Luo, 9 Sep 2025).

The same paper introduces a categorical Hodge correspondence by enhancing Saito’s derived category of mixed Hodge modules to a stable $H^3(X,\mathcal O_X)$1-category $H^3(X,\mathcal O_X)$2, whose heart recovers $H^3(X,\mathcal O_X)$3. Taking $H^3(X,\mathcal O_X)$4-theory yields

$H^3(X,\mathcal O_X)$5

together with a Chern character

$H^3(X,\mathcal O_X)$6

The obstruction formalism extends to this categorical setting through an exact triangle of categories or a long exact sequence in $H^3(X,\mathcal O_X)$7-theory measuring failure of lifting from $H^3(X,\mathcal O_X)$8 to $H^3(X,\mathcal O_X)$9 (Luo, 9 Sep 2025).

In this framework, obstruction theory is primarily an extension theory across the boundary. This suggests a structural link between boundary geometry, mixed Hodge modules, and the detectability of Hodge classes by compactification data.

4. $H^2_D(X,\mathcal O_X)$0-adic liftability and Tate-class obstructions

A different obstruction theory concerns liftability from positive characteristic. Let $H^2_D(X,\mathcal O_X)$1 be smooth projective over a number field, $H^2_D(X,\mathcal O_X)$2 a prime of good reduction, and $H^2_D(X,\mathcal O_X)$3 the reduction. The relevant cohomology theories are de Rham, $H^2_D(X,\mathcal O_X)$4-adic, and crystalline cohomology, together with their cycle-class maps. Crystalline Tate classes are defined by

$H^2_D(X,\mathcal O_X)$5

Through the Berthelot–Shiho comparison isomorphism, the Hodge filtration on de Rham cohomology induces a filtration $H^2_D(X,\mathcal O_X)$6 on crystalline cohomology (Costa et al., 2020).

The key input is the theorem of Bloch–Esnault–Kerz: if $H^2_D(X,\mathcal O_X)$7 is smooth projective and $H^2_D(X,\mathcal O_X)$8, then an integral cycle $H^2_D(X,\mathcal O_X)$9 lifts to $G=\mathrm{SO}(V,Q)$0 if and only if its crystalline class lies in $G=\mathrm{SO}(V,Q)$1. Equivalently, for the quotient map

$G=\mathrm{SO}(V,Q)$2

one has

$G=\mathrm{SO}(V,Q)$3

Restricting $G=\mathrm{SO}(V,Q)$4 to the Tate subspace yields the primary obstruction

$G=\mathrm{SO}(V,Q)$5

where $G=\mathrm{SO}(V,Q)$6 (Costa et al., 2020).

This obstruction gives the bound

$G=\mathrm{SO}(V,Q)$7

A finer analysis decomposes $G=\mathrm{SO}(V,Q)$8 using cyclotomic factors of the characteristic polynomial of $G=\mathrm{SO}(V,Q)$9, producing higher obstruction maps $\dim_{\mathbb Q}V=28$0 on the summands $\dim_{\mathbb Q}V=28$1. Under the full Tate conjecture, one gets the stronger estimate

$\dim_{\mathbb Q}V=28$2

(Costa et al., 2020).

The paper also gives an effective algorithm for hypersurfaces. It computes a basis of de Rham cohomology respecting the Hodge filtration, approximates Frobenius $\dim_{\mathbb Q}V=28$3-adically using Monsky–Washnitzer or controlled reduction, reconstructs the characteristic polynomial, factors it into cyclotomic pieces, computes approximate Tate subspaces, and then evaluates the obstruction maps to obtain rigorous upper bounds on $\dim_{\mathbb Q}V=28$4 from a single prime reduction (Costa et al., 2020). The examples show dimension drops by $\dim_{\mathbb Q}V=28$5 or more at one prime, in contrast with classical two-prime methods.

In this arithmetic setting, obstruction theory does not primarily test algebraicity of a complex Hodge class. It tests whether a class on the special fiber can be lifted through the Hodge filtration. A plausible implication is that filtered $\dim_{\mathbb Q}V=28$6-module structure encodes a more refined obstruction than the mere existence of Frobenius-fixed classes.

5. Local deformation obstructions for divisor classes

For a smooth projective variety $\dim_{\mathbb Q}V=28$7 and an effective Cartier divisor $\dim_{\mathbb Q}V=28$8, the classical topological obstruction to deforming the fundamental class of $\dim_{\mathbb Q}V=28$9 as a Hodge class lies in $(p,p)$00. A refinement replaces this with a local cohomology group

$(p,p)$01

yielding a local topological obstruction theory (Biswas et al., 2020).

The divisor defines a local class

$(p,p)$02

Contraction against $(p,p)$03 produces maps

$(p,p)$04

which fit into a diagram with the classical cup-product obstruction in $(p,p)$05 (Biswas et al., 2020).

For a family $(p,p)$06 with Kodaira–Spencer map $(p,p)$07, the local topological obstruction attached to $(p,p)$08 is

$(p,p)$09

The key criterion is: $(p,p)$10 In particular, vanishing of $(p,p)$11 implies that $(p,p)$12 remains of type $(p,p)$13 in the deformation (Biswas et al., 2020).

This obstruction is strictly finer than the classical one, because the edge map

$(p,p)$14

need not be injective. The paper proves that after modding out by a certain image from $(p,p)$15, vanishing of the local obstruction is equivalent to vanishing of the classical cup-product obstruction, leading to the notion of $(p,p)$16-semi-regularity (Biswas et al., 2020). It also develops a comparison with geometric obstruction theory: the obstruction to lifting $(p,p)$17 as an effective Cartier divisor lies in $(p,p)$18, and

$(p,p)$19

The distinction between deformation of the cohomology class and deformation of the divisor itself is central. The paper constructs first-order deformations in which $(p,p)$20 deforms as a Hodge $(p,p)$21-class while $(p,p)$22 does not lift as an effective Cartier divisor (Biswas et al., 2020). This sharply separates Hodge-theoretic persistence from geometric realizability.

6. Integral, torsion, and non-algebraicity phenomena

Integral Hodge classes introduce a further layer of obstruction because torsion and coefficient exact sequences become essential. A recent construction starts from Enriques surfaces $(p,p)$23, with

$(p,p)$24

the K3 double-cover class and, for $(p,p)$25,

$(p,p)$26

Brauer-detecting classes. Their external cup-product is

$(p,p)$27

where $(p,p)$28. The coefficient sequence

$(p,p)$29

yields the Bockstein class

$(p,p)$30

Each $(p,p)$31 is a priori a $(p,p)$32-torsion Hodge class in codimension $(p,p)$33 (Rahman, 4 May 2026).

To detect non-algebraicity in higher degree, the construction uses the MacPherson–Vilonen obstruction-channel formalism. External tensor products on perverse sheaves, a categorical Bockstein boundary, and a Leibniz rule for the MV boundary isolate a distinguished Enriques–Brauer component

$(p,p)$34

The MV obstruction-channel image of $(p,p)$35 has nonzero projection to this component (Rahman, 4 May 2026).

The algebraicity test is separated from the detection of nonzero obstruction by the Brauer-separation hypothesis, which asserts that algebraic codimension-$(p,p)$36 cycle classes have zero projection to the same Enriques–Brauer component. Decomposable cycles are shown to vanish under this projection, and the remaining non-decomposable case is reduced using degree-$(p,p)$37 even Chow–Künneth projectors on the Enriques factors to a single finite-coefficient control problem in the $(p,p)$38 direction (Rahman, 4 May 2026).

Under the Brauer-separation hypothesis, $(p,p)$39 is a non-algebraic $(p,p)$40-torsion integral Hodge class. In the degree-$(p,p)$41 case, this reinterprets Diaz’s Enriques-product construction as the level-two instance of an $(p,p)$42-fold cup-product Bockstein mechanism (Rahman, 4 May 2026).

This setting emphasizes that obstruction theory for Hodge classes is not confined to rational classes. Integral and torsion phenomena require coefficient-level constructions, Bockstein boundaries, and refined tests beyond standard Hodge-type considerations.

7. Synthesis and current implications

The recent literature exhibits several non-equivalent meanings of “obstruction theory for Hodge classes.”

Setting	Obstruction object	Question tested
Shimura varieties	Non-interiority of $(p,p)$43	Can known cycle constructions realize a given Hodge class?
Toroidal varieties	$(p,p)$44	Does a Hodge class extend across the boundary?
Positive characteristic	$(p,p)$45 and higher $(p,p)$46	Does a Tate class lift through the Hodge filtration?
Divisor deformations	$(p,p)$47	Does a divisor class or some multiple deform geometrically?
Integral Hodge theory	Bockstein/MV obstruction channel	Is a torsion Hodge class algebraic?

A common theme is that the Hodge condition alone is too coarse. In every setting above, one begins with a class having the expected Hodge-theoretic type and then introduces an auxiliary structure—boundary behavior, automorphic spectrum, filtered Frobenius, local cohomology, or coefficient exact sequences—to determine whether the class can be realized in a stronger sense.

The automorphic example on $(p,p)$48 is especially notable because it gives an explicit rational Hodge class that is unconditionally constructed yet lies outside all known geometric constructions on that Shimura variety (Mostaed, 6 Feb 2026). The toroidal theory, by contrast, identifies a vanishing criterion under which boundary obstruction disappears and extension holds (Luo, 9 Sep 2025). The $(p,p)$49-adic theory converts obstruction into an effective computational tool (Costa et al., 2020). The divisor case isolates the gap between deformation as a Hodge class and deformation as a divisor (Biswas et al., 2020). The Enriques-product constructions show that integral Hodge counterexamples can be organized by a systematic Bockstein mechanism (Rahman, 4 May 2026).

Taken together, these developments indicate that obstruction theory for Hodge classes has become a family of techniques rather than a single framework. A plausible implication is that future progress will depend on relating these frameworks more directly: spectral obstructions in automorphic cohomology, boundary obstructions in log geometry, and torsion or filtered obstructions in arithmetic geometry may encode different aspects of the same gap between cohomological existence and algebraic realization.