The central goal of this paper is to prove a "group chunk theorem" in a very general mathematical setting. A group chunk is roughly a partial object that behaves like a group on a dense core but may not be fully defined everywhere. The classical theorems of this kind, appearing separately in model theory (a branch of mathematical logic) and algebraic geometry (the study of geometric objects defined by polynomial equations), say that such partial group-like structures can always be completed into genuine groups. The authors unify these scattered results by working in the language of sheaves on sites, a flexible framework borrowed from modern algebraic geometry that can encode both logical and geometric settings at once.
The second main contribution adapts a technique due to the logician Ehud Hrushovski, who showed how to extract a group chunk from subtle symmetry data called "germs of definable functions on stationary types." In the original model-theoretic version, this relied on logical tools such as canonical bases and elimination of imaginaries. The authors replace these with a purely geometric substitute: Hilbert schemes, which are spaces that parametrize families of geometric subobjects in a systematic way. This substitution lets the result work over much more general base spaces than were previously accessible, broadening its applicability in algebraic geometry.
Along the way, the paper develops several technical tools that are likely useful beyond this specific project. These include a general theory of "partial morphisms" in abstract categories, a result showing that certain logically defined quotient sets can be understood as sheaf quotients, and a careful treatment of rational maps between schemes over an arbitrary base. Taken together, the paper builds a bridge between model theory and algebraic geometry at a deeper level than before, showing that results which looked specific to each field are actually instances of one unified phenomenon.