A classical Torelli theorem asks whether a geometric object (like a hypersurface, meaning the set of solutions to a single polynomial equation) can be uniquely recovered from certain algebraic data associated to it. The "categorical" version of this question replaces traditional algebraic invariants with a category, which is a mathematical structure capturing objects and the relationships between them. In recent decades, mathematicians have found that a geometric space can often be reconstructed from a special category built out of it, and proving that this reconstruction works is called a categorical Torelli theorem. This paper establishes such theorems for a broad class of hypersurfaces, including both ordinary ones (defined by a single homogeneous polynomial) and weighted ones (where the variables carry different scaling weights).
The key technical tool is the singularity category, which is a category that keeps track of how badly behaved or "singular" a space is near its worst points. The authors connect this to the Kuznetsov component (a piece of the derived category used for Fano hypersurfaces, which curve inward) and to graded matrix factorizations (algebraic objects used for hypersurfaces of general type, which curve outward). A crucial observation is that under a mild numerical condition on the degree and weights, a certain symmetry operation called the internal grading shift can be read off intrinsically from the category itself, rather than being extra data one has to specify by hand. This means the singularity category carries enough information to pin down the geometry.
From this intrinsic category, the authors recover the formal local structure of the singularity using a result called the categorical Mather-Yau theorem. For ordinary hypersurfaces this recovery leads to a projective linear equivalence, meaning you can identify the defining equations up to a change of coordinates. For weighted hypersurfaces, a torus action encodes the different weights and allows recovery of the weighted polynomial itself. The result is a single unified proof that works across different curvature types and generalizes earlier piecemeal results, while also establishing genuinely new theorems for hypersurfaces of general type that were not previously known.