Alec Kriebel

Rational Hodge classes on mixed products of K3 moduli spaces

A Hodge class is a cohomological class with the type expected of an algebraic cycle. The rational Hodge conjecture asks whether every such rational class on a smooth projective complex variety comes from a rational linear combination of algebraic subvarieties. This note records that conclusion for finite mixed products of certain moduli spaces built from projective K3 surfaces.

The result is an assembled corollary of OpenAI's theorem for arbitrary mixed products of K3 surfaces and Bülles's Chow-motive splitting, with Arapura's earlier conditional transfers credited. The note makes the correspondence argument and its hypotheses explicit. The central mixed-K3 theorem belongs to OpenAI; the motivic reduction is inherited.

Original abstract

Combining OpenAI's rational Hodge theorem for arbitrary mixed products of projective K3 surfaces with Bülles' Chow-motive decomposition, we record rational Hodge in every codimension for finite mixed products of the smooth projective K3 moduli spaces covered by that decomposition. The factors may have different K3 bases and include Hilbert schemes, twisted stable-sheaf moduli, and eligible moduli of Bridgeland-stable objects. We give the explicit pull-push argument, retaining smoothness, projectivity, and stability hypotheses. The contribution is an assembled corollary of the cited results; it supplies neither a new proof of the base K3 theorem nor a new motivic reduction.

The precise statement

Let S1, …, Sr be complex projective K3 surfaces, with a Brauer class αi on each Si. Each factor Mi must be an actual smooth projective moduli space in the scope of Bülles's Theorem 0.1:

  • a moduli space of Gieseker-stable αi-twisted sheaves, with the defining sheaf data and polarization fixed; or
  • a moduli space of stable objects in the twisted derived category Db(Si, αi), for a generic Bridgeland stability condition relative to its fixed numerical data, within the cited theorem's scope.

For their finite product X = ∏i=1r Mi, the cycle-class map is surjective in every codimension p:

cl: CHp(X)ℚ → H2p(X, ℚ) ∩ Hp,p(X, ℂ).

The bases may be different or repeated, and the Brauer classes are retained. Smoothness and projectivity of the stable moduli spaces are hypotheses, rather than conclusions for arbitrary numerical data or stability conditions. No additional fine-moduli or primitivity assumption is imposed beyond Bülles's statement.

In particular, this covers every finite mixed product of Hilbert schemes Si[ni] with ni ≥ 0. The case n = 0 is a point, and S[1] = S. An empty factor gives a vacuous assertion. A smooth zero-dimensional projective factor is a finite disjoint union of points, so the argument applies component by component; the product of no factors is a point.

How the transfer works

Bülles's splitting expresses the rational Chow motive of each allowed moduli space as a direct summand of a finite sum of Tate twists of powers of its K3 base. In concrete terms, algebraic correspondences send classes to those powers and back, with the sum of the return compositions equal to the diagonal.

Tensoring the individual splittings produces intermediate varieties KJ that are mixed products of powers of the possibly different K3 bases. OpenAI's mixed-product theorem supplies rational Hodge for each KJ. The argument applies to arbitrary product Hodge classes; it does not presume that they decompose into products of classes from the individual factors.

The degree bookkeeping is explicit. Write d = dim X and ej = dim Kj. The split correspondences have codimensions aj and bj, with

aj + bj = d + ej,   ∑j Bj ∘ Aj = ΔX.

A codimension-p Hodge class is sent to codimension qj = p + aj − d. Choose an algebraic representative there and apply the return correspondence. Its codimension is

qj + bj − ej = p.

Summing the returned cycles recovers the original class by the diagonal identity. Targets outside the cohomological range are zero. Using actual cycle codimensions avoids ambiguity from Tate-twist sign conventions.

The non-fine and twisted cases use Bülles's actual rational Chow correspondences and the normalized diagonal construction underlying Markman's result. They do not require an honest universal family. Raw duplication of a universal family cannot be substituted for that normalized construction.

Where the mathematical input lies

The OpenAI input is Theorem 1.1 of The rational Hodge conjecture for products of K3 surfaces, read at the pinned public Math Release snapshot adc7f1241b42e322a6451854ab7e4b4c146bf78a. It includes arbitrary distinct or repeated complex projective K3 factors. Separate self-power theorems do not supply this mixed-base premise.

Conversely, taking each Hilbert-scheme factor with n = 1 recovers the mixed-K3 theorem itself. Given Bülles's established splitting, the universal mixed-moduli and mixed-K3 assertions are therefore equivalent. This corollary does not bypass the central imported theorem or provide a new proof of it.

Arapura's earlier work gives conditional transfers to Hilbert schemes and K3 sheaf moduli, and a broader transfer for motivated varieties. Bülles supplies the precise splitting used for the twisted-sheaf and eligible Bridgeland cases. The OpenAI quadratic-locus companion already records a restricted-locus moduli application, and its exact Kuga–Satake criterion together with the later universal theorem supplies the unrestricted-base self-power case. This note records the mixed-moduli assembly, without a claim of firstness or substantial independent novelty. Manuscript cover dates do not establish earliest public priority.

Scope and verification

The conclusion concerns rational coefficients and the actual smooth projective stable moduli spaces specified above. It does not extend the statement to arbitrary deformations of K3[n] type, singular moduli or semistable compactifications, arbitrary abelian-surface moduli, integral or generalized Hodge, or finite-dimensionality of Chow motives.

The package contains the correspondence proof, dependency notes and supporting finite algebra and sign checks. Those computations support their stated finite identities. The upstream Lean Clifford module concerns rational Clifford and tensor identities; it does not formalize algebraic cycles, the geometric Kuga–Satake theorem or this Hodge conclusion. The presentation rests on ordinary mathematical proof review, with no claim of a formal geometric certificate.

Original files and manuscript

  1. Manuscript (PDF, 67,264 bytes) · Zenodo original PDF
  2. Source and verification package (ZIP, 135,014 bytes) · Zenodo original ZIP

Open the four-page PDF · Download PDF · Zenodo record

These are unchanged copies of the original Zenodo files, listed in the record's order. The manuscript and declared publication date are 6 October 2026. The native API metadata does not declare a version.

Credited sources

The manuscript contains the full bibliography and precise theorem interfaces.

Citation and rights

Kriebel, Alec. (2026). Rational Hodge classes on mixed products of K3 moduli spaces [Unrefereed preprint]. Zenodo. doi:10.5281/zenodo.23204176. Download BibTeX.

The original text, source, code, notes and recorded results are licensed CC BY 4.0. Cited third-party works retain their authorship and rights.

Status and collaboration

AI tools were used extensively in research, source checks, drafting and verification. This is an unrefereed, AI-assisted preprint; automated adversarial reviews are not conventional human peer review. The author is responsible for the claims, and the OpenAI inputs are attributed to OpenAI.

Substantive independent verification of the proof dependencies and correspondence argument, improvements to the exposition, and development of a journal version would be useful contributions. Collaboration and authorship would be based on actual contributions.