How can quantum connection exponents prove non-rationality of threefolds and fourfolds?

Two numbers read off a generic Hodge factor

The object being measured is not the whole cohomology of the variety but a single differential module attached to it. For a smooth complex projective variety X, the A-model F-bundle carries a meromorphic flat connection on B_X × D, and reducing it at u = 0 produces the residual endomorphism K^F. Its generalized eigenspaces split the Hodge-invariant part of cohomology into generic Hodge factors M_{λ,b}, indexed by the eigenvalues λ of K^F.

A factor is called cyclic when the restriction of K^F to the corresponding generalized eigenspace has a single Jordan block. When such a factor is of unramified exponential type, it factors as E^{-λ/u} ⊗ R_{λ,b} with R_{λ,b} regular singular. The two invariants are then the rank m of the module and the multiset F = Exp(R_{λ,b}) of negative residue eigenvalues modulo Z, counted with algebraic multiplicity. The authors state explicitly that these are not invariants of Hodge atoms, so the reader should not conflate the two notions.

Rational varieties allow only three exponent pairs

Theorem 4.5 is the criterion. Let X be a connected smooth complex projective variety of dimension 2 ≤ n ≤ 4. Suppose a generic Hodge factor has a cyclic representative of unramified exponential type with rational residues, with rank m and regularized exponent multiset F. If X is rational, then (m, F) must be one of three entries: rank 1 with F = {{0}} (the point type), rank 2 with F = {{1/2, 1/2}} (the curve type), or rank 3 with F = {{0, 0, 0}} (the surface type).

The logic is a shortlist, not a classification of rational varieties. The theorem says what a rational X of dimension at most four is allowed to produce; it does not say that every variety producing one of these pairs is rational. The contrapositive is what carries the applications: an admissible factor whose (m, F) falls outside the three rows obstructs rationality.

Persistence through blowups and projective bundles

The criterion is only useful if the invariant does not change under the operations that relate birationally equivalent varieties. The authors establish this in two steps. Proposition 4.2 shows that if one connected component of the generic spectral cover has a single k-rational cyclic point where the selected factor is of unramified exponential type with rational exponents, then every k-rational cyclic point on that component shares the same properties and the same exponent multiset.

Lemma 4.3 then shows that a selected factor meeting these hypotheses can be followed through blowup and projective-bundle comparisons while preserving its rank, cyclicity, and exponents. These results rest on the formal-persistence statement recorded in the paper as Theorem A.3, whose proof tracks the regular-singular factor along the flow and shows the transported summand keeps the same constant residue matrix. Because the invariant is stable along these comparisons, a mismatch with the three-row shortlist cannot be removed by passing to a birationally equivalent model of the kind considered.

Cubic threefolds, Gushel–Mukai threefolds, quartic fourfolds

Section 5 turns the criterion into concrete conclusions. For cubic threefolds the authors recover the irrationality theorem of Clemens–Griffiths, and they further prove that Y × P1 is irrational for every smooth cubic threefold Y, a conclusion they note was also established by Gong–Xu. For Gushel–Mukai threefolds, where Beauville proved irrationality of a general member and Debarre–Mongardi constructed explicit irrational examples, the criterion here proves irrationality of every smooth ordinary Gushel–Mukai threefold and of its product with P1.

For very general quartic fourfolds, where Totaro proved stable irrationality, the criterion recovers the irrationality conclusion. The scope of each statement matters: the Gushel–Mukai result is stated for every smooth ordinary member, while the quartic fourfold result is stated for very general members, not for every smooth quartic fourfold. The authors also note that a catalog of small quantum multiplication matrices and a software package for computing them were built, and that these will be updated to compute the exponent multiset for many smooth projective threefolds.

A conjectural link to the upper Serre dimension

The paper opens with a motivating question about the residual category A_Y of a smooth Fano complete intersection Y of dimension m, whose upper Serre dimension is σ(Y) = m − 2r_Y/d_r, where r_Y is the Fano index. The question asks whether σ(Y) > m − 2, equivalently r_Y < d_r, forces irrationality for a very general member, and in which cases irrationality holds for every smooth member.

The authors do not claim to have settled this question. What they report is a numerical match in the computed examples. For smooth cubic threefolds, the representatives mod Z for the eigenvalue 0 are s = −5/6 and −1/6, and −5/6 equals −1/2 times the upper Serre dimension for the Kuznetsov component of the semiorthogonal decomposition. The same relation is reported for the smooth quartic fourfolds computed, and for Gushel–Mukai threefolds at the eigenvalue 6q with q a Novikov variable, where the corresponding upper Serre dimension is computed elsewhere. The authors describe the relation between σ(Y) and the exponents F as expected, not established.

Dimension four and the cyclicity hypothesis

Theorem 4.5 is stated for dimensions 2 ≤ n ≤ 4, so the three-row shortlist is not asserted for higher-dimensional varieties. The criterion also requires the generic Hodge factor to have a cyclic representative of unramified exponential type with rational residues; a factor failing cyclicity or the unramified exponential type condition falls outside the theorem as stated. The persistence argument in Lemma 4.3 covers blowup and projective-bundle comparisons, which is the class of birational comparisons the paper considers.

One further boundary is worth keeping in view when reading the applications. The authors note that during preparation they learned Andrew Harder, Jiaji Cai, and Sukjoo Lee were independently working out the quartic fourfold example by similar methods, and they credit Cai with an obstruction obtained from the formal monodromy of the quantum connection. The criterion is therefore one route among several to overlapping conclusions, not the only available obstruction.

The exponent multiset is the quantity to compute next

The paper's own forward-looking priority is computational. The authors state that the catalog and software package will be updated to compute the exponent multiset for many smooth projective threefolds, which would yield further applications of Theorem 4.5. They give one anticipated example: the smooth complete intersection of a quadric and a cubic in P5 is expected to be shown irrational via Theorem 4.5 together with that computation, offering an alternative to earlier arguments.

That expectation is stated as an anticipation, not as a completed result, and it depends on a computation not yet reported in the paper. The practical reading is therefore narrow and specific: for a smooth projective variety of dimension 2 to 4 admitting a cyclic generic Hodge factor of unramified exponential type with rational residues, compute the rank and the regularized exponent multiset, compare the pair against the three rational rows, and treat a mismatch as a non-rationality certificate. The most consequential open item is whether the exponent multiset can be computed broadly enough to extend the same test beyond the examples treated here.

Sources used in this answer