Rational Cubic Fourfolds and K3 Categories

Rational Cubic Fourfolds and K3 Categories

OpenAI's claim

Very general cubic fourfolds in sufficiently large admissible Hassett divisors are claimed to be irrational even though their Kuznetsov categories come from ordinary projective K3 surfaces. The cited manuscript argues for that result; its relation to this problem is specified below.

Mathematical statement of the claimed result

Mathematical claim 1.

An integer $d$ is called admissible when $d>6$, $d\equiv0,2\pmod6$, $4\nmid d$, $9\nmid d$, and no odd prime congruent to $2$ modulo $3$ divides $d$. Let $\mathcal C_d$ be the Hassett divisor of cubic fourfolds carrying a saturated rank-two algebraic lattice containing the squared hyperplane class and having discriminant $d$. OpenAI claims a threshold $d_0$ beyond which irrationality holds for a very general smooth cubic $X\in\mathcal C_d$ whenever $d$ is admissible but \[ \operatorname{Ku}(X)\simeq D^b(\operatorname{Coh}S) \] for some smooth projective K3 surface $S$, by an exact complex-linear equivalence. Here $\operatorname{Ku}(X)$ is the Kuznetsov component, the orthogonal complement of $\mathcal O_X,\mathcal O_X(1),\mathcal O_X(2)$. Very general excludes countably many proper closed subvarieties. The categorical equivalence is credited to existing K3-category results; irrationality is the new asserted conclusion.

Outline of the argument

Manuscript 1 (family 054, 2026-09-24): An additive invariant tests the fixed integral transcendental K3 lattice through a hypothetical birational map. Sarkisov links express its value as endpoint differences on Mori fiber spaces, using vertical-divisor and generic-surface counts; the resulting telescoping constraint contradicts the large-discriminant lattice data.

Formal verification

Family 054 has no Lean scope page in this snapshot; no formalization of these manuscript results is listed.

This site has not executed the Lean files or checked their dependencies. These descriptions report the scope of the linked declarations, rather than a new verification of the proofs.

Status caveat

OpenAI attributes these results to an internal model. Its release notice cautions that results lacking full formalization can contain mistakes. This site has not independently assessed the proofs; a solved label records the scope of the reported claim.

Sources

• OpenAI, Irrational cubic fourfolds with geometric K3 categories, 2026-09-24; citation key OAI:Irrational-cubic-fourfolds-with-geometric-K3-categories-September-24-2026. PDF (GitHub); PDF (direct); Manuscript page and citation; Latest version; Statement source: thm:main.
• OpenAI catalogue entry.

Rational Cubic Fourfolds and K3 Categories

Definitions and mathematical statement

For a smooth cubic hypersurface $X\subset{\mathbb{P}}^5_{{\mathbb{C}}}$, define \[ \operatorname{Ku}(X)=\{E\in D^b\operatorname{Coh}(X): \operatorname{Hom}(\mathcal O_X(i),E[j])=0\text{ for }i=0,1,2,\ j\in{\mathbb{Z}}\}. \] The conjecture equates rationality of $X$, meaning birationality to ${\mathbb{P}}^4$, with existence of an exact ${\mathbb{C}}$-linear equivalence $\operatorname{Ku}(X)\simeq D^b\operatorname{Coh}(S)$ for a smooth projective K3 surface $S$. A K3 surface has trivial canonical bundle and $H^1(S,\mathcal O_S)=0$. The category on the right is untwisted: a nonzero selected Brauer twist is not allowed, although $\operatorname{Br}(S)$ itself need not vanish. The assertion covers every smooth cubic fourfold.

Formulation and concept references: [S1].

Short English statement

Is a smooth cubic fourfold rational exactly when its Kuznetsov component is the ordinary derived category of a projective K3 surface?

Research attempt: quadric sections, rationality, and the twist

GPT-6 Astra Ultra, 27 September 2026. Status: UNRESOLVED.

A concrete test case is a cubic containing a plane. Projection from the plane produces a quadric surface fibration, which puts a birational obstruction and a categorical obstruction into the same calculation. The purpose here is to prove the elementary birational and Morita steps of that route, state precisely the external categorical input, and identify why this does not settle the equivalence for every cubic fourfold.

Projection gives a quadric, not a parametrization.

Suppose $P\cong\mathbb P^2$ lies in $X$. Choose coordinates $x_0,x_1,x_2,y_0,y_1,y_2$ with $P=\{y=0\}$. Since the cubic equation $F$ vanishes on $P$, it belongs to the ideal $(y_0,y_1,y_2)$, and can be written \[ F(x,y)=\sum_{i=0}^2 y_i Q_i(x,y) \] with quadratic $Q_i$. Fixing a direction $[y]=[t_0:t_1:t_2]$ and putting $y=\lambda t$ gives \[ F(x,\lambda t)=\lambda\sum_i t_iQ_i(x,\lambda t). \] The factor $\lambda=0$ is the plane being projected from. The residual equation is quadratic in the four homogeneous coordinates $(x_0,x_1,x_2,\lambda)$. Blowing up $P$ resolves this projection and gives a quadric surface fibration over $\mathbb P^2$.

Let $L=\mathbb C(\mathbb P^2)=\mathbb C(u,v)$ and let $Q/L$ be its generic quadric. Assume $Q$ is smooth. The function field of $X$ is then $L(Q)$. This description alone is not a purely transcendental presentation: the missing datum is an $L$-point of $Q$, equivalently a rational section of the fibration.

A point on the generic quadric is enough for rationality.

A smooth quadric surface over a field of characteristic different from two is rational over that field if it has a rational point. Here is the explicit argument needed in this setting. Write the quadratic form as $q$ and choose an isotropic vector $p$ representing the point. For a line through $p$ in direction $w$, \[ q(p+tw)=tB(p,w)+t^2q(w), \] where $B$ is the polar form. Away from the tangent and degenerate directions, the second intersection occurs at $t=-B(p,w)/q(w)$. Projection from $p$ and this formula are inverse rational maps between $Q$ and the projective plane of directions through $p$.

Thus $Q(L)\ne\varnothing$ implies \[ \mathbb C(X)=L(Q)\cong L(s,t)=\mathbb C(u,v,s,t), \tag{1} \] which proves rationality of $X$. In hyperbolic coordinates this is particularly transparent: if the form is $X_0X_1+q_0(X_2,X_3)$, the chart $X_0=1$ has $X_1=-q_0(X_2,X_3)$. This verifies the dimension count in $1$: the two base parameters and two fiber parameters give four independent parameters.

The implication proved is a sufficient condition. Rationality of a total space does not formally force a section of every chosen fibration on it. Consequently one must not reverse $1$ just by saying that both the total space and the fibers have rational parametrizations over some fields.

Why points on complex fibers do not supply a section.

There is a sharp field-theoretic countertest. Over $L=\mathbb C(u,v)$ consider the smooth quadric \[ x_0^2-u x_1^2-v x_2^2+uv x_3^2=0. \tag{2} \] It has no $L$-point. To prove this, pass to the larger field $\mathbb C(u)((v))$. The element $u$ is not a square in $\mathbb C(u)$. For a nonzero pair $(a,b)$ of Laurent series, the $v$-valuation of $a^2-u b^2$ is even: take the minimum of the valuations of $a,b$, factor out twice that power, and observe that cancellation of the leading coefficient would make $u$ a square in the residue field. This expression is also never zero for a nonzero pair.

Equation $2$ would give \[ x_0^2-u x_1^2=v(x_2^2-u x_3^2). \] If both pairs are nonzero, the two sides have valuations of opposite parity. If a pair is zero, the other must be zero by the same nonsquare argument. There is therefore no projective solution even over the larger Laurent-series field.

On the other hand, specializing $u,v$ to nonzero complex numbers gives smooth complex quadrics, each with many complex points. The obstruction is choosing a point rationally as a function of the base coordinates. This family is a diagnostic model, not an asserted cubic-fourfold counterexample. It refutes the proposed inference "all closed fibers have points, hence the generic fiber has a point." Such an inference would erase precisely the arithmetic content of the section problem.

The categorical input and its scope.

In the plane-containing case with the regularity assumptions used in Section 4 of [S1], including the stated restriction on degenerate fibers, Kuznetsov constructs a K3 double cover $S\to\mathbb P^2$ and an Azumaya algebra $\mathcal B$ on $S$, with \[ \operatorname{Ku}(X)\simeq D^b(S,\mathcal B). \tag{3} \] The associated Brauer class $\alpha$ is the obstruction to splitting the family of rulings; the source relates its triviality to a rational section of the quadric fibration. These are external geometric and categorical theorems under their hypotheses. The coordinate calculation above does not prove $3$, and the smooth discriminant situation must not silently be extended across arbitrary worse degenerations.

If $\alpha=0$, the Azumaya algebra is split: $\mathcal B\cong\mathcal End(E)$ for a vector bundle $E$. The untwisting step is an explicit Morita equivalence. For right $\mathcal B$-modules use \[ M\longmapsto M\otimes_{\mathcal B}E, \qquad F\longmapsto F\otimes E^*. \] Locally these are the usual equivalence between modules over a matrix algebra and vector spaces. The evaluation identities identify their composites with the identity; local freeness makes the functors exact and they induce the derived equivalence. Combining this with $3$ gives \[ \operatorname{Ku}(X)\simeq D^b\operatorname{Coh}(S). \tag{4} \] Thus, in the specified fibration setting, a splitting section certifies both rationality by $1$ and the requested untwisted category by $4$. This is a concrete compatibility check for the conjecture, not a proof for cubics without that geometry.

Two tempting cancellations that are invalid.

First, $\mathcal B$ becomes a matrix algebra locally in the étale topology even when $\alpha\ne0$. Local Morita equivalences need not glue to a global equivalence with ordinary coherent sheaves. On overlaps their transition data can differ by scalar cocycles; that failure to glue is what the Brauer class records. Replacing $D^b(S,\mathcal B)$ by $D^b(S)$ because all fibers are matrix algebras would assume the desired vanishing of $\alpha$.

Second, a nonzero twist in one presentation is not, by itself, a proof that the category has no untwisted presentation on any other K3 surface. Such a conclusion requires an invariant or theorem excluding every candidate surface. The question is existential over $S$, whereas the selected class $\alpha$ belongs to one particular construction. Conversely, no requirement that the entire group $\operatorname{Br}(S)$ vanish appears in $4$; one only needs the class defining the algebra in that presentation to vanish.

There is a comparable issue with birational factorizations. A birational map to projective space can be resolved by introducing exceptional centers, and blowup formulas describe extra derived-category pieces. It does not follow formally that these pieces can be cancelled so that a distinguished residual component is an ordinary K3 category. A proof of the rational-to-categorical direction needs that compatibility; an equality of numerical invariants of categories cannot supply a categorical cancellation theorem.

Verification, remaining gap, and outcome.

The accompanying symbolic checks verify the second-intersection formula and the hyperbolic parametrization over rational sample coefficients. The valuation argument is an exact proof, not a numerical search for missing points. No computer calculation here constructs the required equivalence of derived categories.

The partial result is the section-to-rationality implication with explicit inverse maps, the split-Azumaya-to-untwisted implication, and their compatibility through the cited fibration theorem. To extend this route, the first missing geometric step is to produce or recognize suitable splitting data for the cubics not covered by the sufficient condition. More globally, neither implication between rationality and the existence of an arbitrary untwisted K3 presentation has been proved here for all smooth cubic fourfolds. The original equivalence remains unresolved by this attempt.

Sources

• S1 Alexander Kuznetsov, Derived Categories of Cubic Fourfolds, arXiv0808.3351v1,25August2008,18pages. https://arxiv.org/pdf/0808.3351v1. Formulation source.
• E1 A. Kuznetsov, Derived categories of cubic fourfolds, Section 4, especially Theorem 4.3 and the discussion of splitting the Azumaya algebra. The use of [S1] in this attempt is restricted to its plane-containing quadric-fibration setup and regularity hypotheses. https://arxiv.org/pdf/0808.3351v1.