Lattice polytopes are geometric shapes whose corners all sit on grid points (points with whole number coordinates). Mathematicians have developed ways to measure properties of these shapes, including something called the h*-vector, which encodes information about how many grid points the polytope and its scaled-up versions contain. A long-standing open question is whether certain well-behaved polytopes always have h*-vectors that are "unimodal," meaning the sequence of numbers rises to a peak and then falls, without any dips or bumps along the way.
The paper focuses on a class of polytopes called "very ample," which satisfy a decomposition property slightly weaker than the gold standard known as the integer decomposition property (IDP). Roughly speaking, very ample means that if you look at a large enough scaled version of the polytope, every interior grid point can be broken down as a sum of grid points from the original shape. Researchers had wondered whether being very ample was enough to guarantee a unimodal h*-vector, as this would suggest a deep connection between geometric regularity and combinatorial structure.
The authors answer this question by constructing a specific counterexample: a very ample polytope whose h*-vector is not unimodal, meaning the sequence goes up, dips, and rises again. The example is built by taking the Cartesian square of a polytope from a previously known family of very ample constructions. Notably, the counterexample was discovered with help from an AI tool, ChatGPT. The harder and still unresolved question is whether the stronger IDP condition guarantees unimodality, and this paper narrows the landscape by ruling out the weaker condition as sufficient.