The paper deals with a deep question in algebraic geometry: when can a high-dimensional geometric shape called a fourfold be "rational," meaning it can be parametrized in a simple, essentially coordinate-like way? This question is notoriously difficult, and the paper builds on recent work that introduced a new tool, a birational invariant, for detecting irrationality in certain fourfolds called cubic fourfolds (the zero sets of degree-three polynomial equations in five-dimensional projective space). The paper is situated at the intersection of Hodge theory, which studies how complex geometry is encoded in algebraic data about cohomology (a way of measuring the shape of a space), and derived category methods involving so-called "atoms."
The main technical contributions involve showing that a formula due to Iritani, which describes how cohomology behaves under a geometric operation called a blow-up, is compatible with classical decompositions of integral Hodge structures. This compatibility is important because it allows the authors to reinterpret the recently introduced irrationality invariant in terms of atoms, which are building blocks arising from a framework developed by Katzarkov, Kontsevich, Pantev, and Yu. The authors also prove a precise structural result: for every smooth rational cubic fourfold, a specific piece of its cohomology called the primitive cohomology is isometric, as an integral Hodge structure, to the twisted middle cohomology of a simpler and well-understood object called a K3 surface.
As a concrete application, the paper proves that a certain class of fourfolds called Verra fourfolds, in their most general form, are irrational. Verra fourfolds are defined by specific geometric constructions and have been studied as analogues of cubic fourfolds. The irrationality result for very general Verra fourfolds is a new and meaningful advance, since irrationality results for fourfolds remain rare and hard to establish. The paper thus connects abstract Hodge-theoretic machinery to concrete geometric conclusions about which fourfolds cannot be simply parametrized.