gpt 6 astra pro attempt on Math-30

Grothendieck's Variational Hodge Conjecture

Definitions and mathematical statement

For a smooth projective family $f:X\to S$, the relative de Rham bundle $\mathcal H^{2p}_{\mathrm{dR}}=\mathbf R^{2p}f_*\Omega^\bullet_{X/S}$ has the Gauss--Manin connection $\nabla$. A global algebraic section $\xi$ is horizontal when $\nabla\xi=0$, so analytically it is transported by the cohomological local system. Let $\mathrm{cl}_{\mathrm{dR}}$ be the cycle-class map from rational codimension-$p$ cycles on a fiber. The target is \[ \xi(s_0)\in\operatorname{im}\mathrm{cl}_{\mathrm{dR},s_0} \quad\Longrightarrow\quad \xi(s)\in\operatorname{im}\mathrm{cl}_{\mathrm{dR},s}\quad\text{for all }s\in S({\mathbb{C}}). \] Here $S$ is smooth and connected. No single relative cycle realizing all fibers is demanded; only fiberwise algebraicity of the transported class is asserted.

Formulation and concept references: [S1].

Short English statement

If an algebraic cohomology class is transported horizontally through a smooth projective family, must it remain algebraic on every fiber?

Research attempt

Current frontier.

The exact global formulation matters. Charles--Schnell's Theorem 33 and Proposition 34 show that a global flat section which is a rational Hodge class at one point remains Hodge everywhere, and that the Hodge conjecture would imply this variational statement [S1]. Thus an analytic disk on which a locally transported class leaves its Hodge locus is not a counterexample to the printed global algebraic formulation. Recent relevant work includes Pridham's 2024 revision on semiregularity and obstruction maps [E1], and Bae--Kool--Park's 2026 publication giving a variational Hodge application for Calabi--Yau fourfold families with a nonzero reduced virtual cycle [E3]. Such extra geometric or nonvanishing conditions are not available for an arbitrary input here.

Attack route.

The Hodge-conjecture route imports a larger unresolved statement; deformation theory must remove actual cycle obstructions, not merely preserve Hodge type. We instead parameterize representatives by proper Hilbert schemes and ask whether one parameter component dominates the base. We then test whether first-order semiregularity can supply this missing dominance.

Attempt: rationality and the correct local reduction.

Via the comparison isomorphism, with the Tate twist consistent with the cycle-class normalization, $\xi$ corresponds to a global section \[ \alpha\in H^0(S^{\mathrm{an}},R^{2p}f_*{\mathbb{Q}}(p)). \] Indeed, flatness gives a complex local-system section and rationality at $s_0$ propagates along paths. The fixed-part conclusion from the cited invariant-cycle theorem gives the Hodge condition throughout the base [S1], Propositions 32 and 34.

We first work over a nonempty affine open in $S$, so that projective embeddings and the quoted quasi-projective theorems apply. This is not a permanent restriction of the target. A smooth connected complex algebraic variety is irreducible, so any two nonempty affine opens intersect. A universal result on affine bases, started on one containing $s_0$, would therefore propagate across an affine cover by taking an algebraic fiber class at an overlap as the next starting point. The relative dimension $n$ is constant. For $p>n$ the target class is zero; below assume $0\le p\le n$.

Attempt: a countable proper parameterization.

Fix a relatively very ample line bundle. For finite data \[ m\ge1,\quad r\ge1,\quad a_1,\ldots,a_r\in{\mathbb{Z}}, \quad P_1,\ldots,P_r\text{ Hilbert polynomials of degree }n-p, \] consider \begin{equation}\label{wsref-0-65-71-31-35-37-2d-68-69-6c-62-65-72-74} T=\prod_{i=1}^r{}_S\operatorname{Hilb}_{P_i}(X/S). \end{equation} This is a projective finite-type $S$-scheme, with universal flat closed subschemes $Z_i\subset X\times_ST$; see the Hilbert case of [E2], Theorem 5.1. Take their fundamental cycles of relative dimension $n-p$, ignoring lower-dimensional embedded contributions. They determine a section \[ \sum_i a_i\operatorname{cl}([Z_{i,t}])-m\alpha_{f_T(t)} \] of the pulled-back rational cohomology local system on $T^{\mathrm{an}}$. Here we use the usual specialization compatibility of fundamental cycle classes in a proper flat family: along a curve in the parameter space, the specialized fundamental cycle has the transported cohomology class. For an irreducible parameter space, curves through a point and a dense open, together with local constancy on the smooth locus, give the same conclusion across its singular points. This explains why flatness of the universal subscheme, rather than smoothness of the Hilbert scheme, is the relevant input.

Restrict to each reduced irreducible component $T_j$ of $T$ on which that section vanishes at some point. It then vanishes identically, since an irreducible complex algebraic variety is analytically path connected. Let $A_j\subseteq S$ be the image of $T_j$. Properness makes $A_j$ Zariski closed. There are only countably many choices of the finite integer and polynomial data, and finitely many components for each choice. Thus the algebraicity locus of this one class has the description \begin{equation}\label{wsref-0-65-71-31-35-37-2d-6c-6f-63-75-73} \mathcal A_\alpha:= \{s:\alpha_s\text{ is a rational algebraic cycle class}\} =\bigcup_{j\ge1}A_j. \end{equation}

Both inclusions deserve checking. A point of $A_j({\mathbb{C}})$ has a complex preimage, whose weighted universal cycles divided by $m$ represent $\alpha_s$. Conversely, an arbitrary rational cycle is a finite integer linear combination of integral subvarieties after multiplying by some $m$. Those subvarieties define points of the corresponding Hilbert schemes, and the containing component gives one of the $A_j$ above. Negative coefficients and varying denominators are indispensable; using only effective cycles of one fixed degree would not parameterize the stated problem.

Baire--properness lemma. If $\mathcal A_\alpha$ is not meagre in $S^{\mathrm{an}}$, then $A_j=S$ for some $j$, and in particular $\mathcal A_\alpha=S({\mathbb{C}})$.

Proof. A proper algebraic closed subset of the smooth irreducible base has empty analytic interior and is nowhere dense. If every $A_j$ were proper, (eq157-locus) would be meagre. This contradicts the hypothesis. Nonempty analytic open sets are not meagre, since complex manifolds are locally compact Baire spaces. $\square$

Increasing finite unions of the $A_j$ are closed, but their union need not be. Membership of $s_0$ supplies no compactness argument for dominance.

Attempt: trace from a dominating parameter space.

There is a stronger useful consequence of $A_j=S$ that can be proved rather than added to the target. Choose a closed point of the generic fiber of $T_j\to S$ and take its reduced closure $T'\subseteq T_j$. The residue field of this point is finite over ${\mathbb{C}}(S)$, so $T'\to S$ is proper, dominant, and generically finite, of some degree $e\ge1$. After shrinking to a nonempty open $U\subset S$, it is finite étale, since the field extension is separable in characteristic zero. Pull back the universal subschemes to $T'$. Their flatness is preserved by base change.

Let $q:X\times_ST'\to X$ be the proper projection. Taking fundamental cycles and proper pushforwards defines \begin{equation}\label{wsref-0-65-71-31-35-37-2d-74-72-61-63-65} Z=\frac1{em}\,q_*\left(\sum_i a_i[Z_i|_{T'}]\right) \in\mathrm{CH}^p(X)_{{\mathbb{Q}}}. \end{equation} The codimension is correct because $T'$ has the same dimension as $S$. For $s\in U({\mathbb{C}})$, the fiber has $e$ preimages, and specialization and proper pushforward give \[ \operatorname{cl}(Z|_{X_s}) =\frac1{em}\sum_{t\in T'_s}\sum_i a_i\operatorname{cl}(Z_{i,t}) =\alpha_s. \] The restriction of a cohomology class on $X$ to the smooth fibers is a global flat section. It follows by connectedness that equality holds on all of $S$. The restriction of the Chow class is interpreted by refined Gysin pullback along $X_s\hookrightarrow X$, a regular embedding because $f$ and $S$ are smooth; flatness of the pushed-forward cycle over all of $S$ is not required.

Thus dominance gives a global rational cycle with the correct fiber classes. This deduction does not assert that a specified cycle at $s_0$ deforms, that one effective flat family works everywhere, or that an integral cycle is obtained. Division by $em$ is legitimate precisely because the target uses rational coefficients.

Attempt: a precise sufficient deformation input.

Consider the following proposed input for each family and class:

After replacing the initial rational cycle by a rationally weighted tuple with the same cohomology class, there are data as in (eq157-hilbert) and a point $t_0$ over $s_0$, with $\sum_i a_i[Z_{i,t_0}]=m\alpha_{s_0}$ in cohomology, such that $T\to S$ is smooth at $t_0$.

If this input holds, the smooth locus containing $t_0$ maps onto a nonempty open subset of $S$. Every sufficiently near parameter point has the same class identity on its relevant component. Its images are contained in $\mathcal A_\alpha$, so the Baire--properness lemma gives $A_j=S$, and (eq157-trace) completes the fiberwise conclusion. This is a conditional implication, not a proof of the quoted input. Formal smoothness for the finite-presentation morphism would suffice; first-order tangent surjectivity would not.

For an lci subvariety $Z\subset X_s$ of codimension $p$, the customary semiregularity map has the form \[ H^1(Z,N_{Z/X_s})\longrightarrow H^{p+1}(X_s,\Omega_{X_s}^{p-1}). \] Pridham explains the annihilation of obstruction classes and its ambient-deformation interpretation, with the more natural truncated de Rham targets needed in full generality [E1], Introduction and Sections 2.4--2.5. That information does not say that the kernel is zero. For a signed tuple, even cancellation of the sum of the cohomological obstruction images need not kill each obstruction to lifting the tuple. The missing bridge is therefore an all-orders unobstructed representative, or another reason that one parameter component dominates. It is not the already available persistence of Hodge type.

Stress test.

A concrete algebraic warning is the finite morphism \[ \operatorname{Spec}{\mathbb{C}}[t]/(t^2)\longrightarrow{\mathbb{A}}^1_{{\mathbb{C}}}, \] induced by the coordinate $t$. At the origin its differential is surjective between one-dimensional tangent spaces. The first-order base arc $t=\varepsilon$ lifts over ${\mathbb{C}}[\varepsilon]/(\varepsilon^2)$, but the same arc cannot lift over ${\mathbb{C}}[\varepsilon]/(\varepsilon^3)$ because its square is nonzero. The image is just the origin. Thus even a proper map with a surjective tangent map can fail completely to dominate the base. This example refutes the proposed tangent-space shortcut, not VHC.

Nor does density of algebraic fibers suffice for the Baire lemma: a countable union of proper closed subsets can be analytically dense. Likewise, the presence of $s_0$ in (eq157-locus) gives only one nonempty closed image. No reason has been established for it to contain an open set. For divisors the Hodge condition plus the Lefschetz $(1,1)$ theorem supplies fiberwise algebraicity, but using that special case in higher codimension would import the Hodge conjecture itself [S1], Proposition 34 and its discussion.

Outcome.

Outcome: Conditional reduction.

The attempt proves a countable-proper-image description of the algebraicity locus, a nonmeagreness-to-dominance criterion, and a trace construction that turns a dominating cycle-parameter component into a global rational cycle with the required fiber classes. This uses standard cycle and Hilbert-scheme tools; no literature priority is claimed.

What remains is a dominating representative component starting from one algebraic fiber, for example via all-orders smooth lifting. Semiregularity alone has not supplied it. The affine-open overlap argument preserves the full target but does not close this gap. No full proof or counterexample is claimed.

Sources

• S1 François Charles and Christian Schnell, Notes on absolute Hodge classes, Conjectures30/31, Proposition32, Theorem33 and Proposition34; base extension by the coordinator's recorded affine-open argument. https://arxiv.org/pdf/1101.3647. Catalogue formulation source.
• E1 J. P. Pridham, Semiregularity as a consequence of Goodwillie's theorem, arXiv:1208.3111v4, 4 November 2024, Introduction and Sections 2.4--2.5. https://arxiv.org/html/1208.3111v4. Consulted 27 September 2026; obstruction annihilation is not used as an injectivity theorem for the semiregularity map.
• E2 Nitin Nitsure, Construction of Hilbert and Quot Schemes, arXiv:math/0504590, 29 April 2005, Section 1 and Theorem 5.1 (Grothendieck). https://arxiv.org/html/math/0504590. Consulted 27 September 2026; representability, universal flat families, and projectivity for fixed Hilbert polynomial.
• E3 Younghan Bae, Martijn Kool, and Hyeonjun Park, Counting surfaces on Calabi--Yau 4-folds I: Foundations, Geometry & Topology 30 $2026$, 2431--2564; arXiv:2208.09474v2, 19 May 2025, abstract's variational Hodge application. https://arxiv.org/abs/2208.09474v2. Consulted 27 September 2026. Used only for the stated restricted frontier result, not as a theorem for arbitrary families.