Model theory is a branch of mathematical logic that studies mathematical structures (like groups, fields, or graphs) through the lens of formal languages and what can be expressed in them. A central concept is "categoricity," which roughly means that a mathematical theory has essentially only one model of a given size. This paper studies a relative version of that idea: instead of asking whether a whole structure is uniquely determined, it asks whether the structure is uniquely determined once you fix a distinguished part of it, called P. The pair of a theory T and its P-part is called a "relatively categorical" theory.
The main contribution is a complete description of what relatively categorical theories look like when T is an "internal cover" of its P-part, meaning the full structure T is tightly controlled by the simpler P-part in a specific technical sense. The authors show that, after some standard bookkeeping moves, these theories always have the same basic shape: you start with the P-part, then attach a new collection of objects that form what is called a torsor (or principal homogeneous space) for some definable group G. A torsor is like a copy of a group that has "forgotten" its identity element, so it looks like the group but has no preferred basepoint. The action of the group on this torsor is the key structure that ties everything together.
The paper also characterizes when T is "relatively omega-stable," a tameness condition that controls how complicated the theory is from a classification standpoint. The answer turns out to depend on a group called the dual binding group H: T is relatively omega-stable precisely when H satisfies a descending chain condition on its definable subgroups, meaning there are no infinite strictly decreasing sequences of such subgroups. This gives a clean algebraic criterion for a logical property, connecting abstract model-theoretic stability theory to concrete group-theoretic conditions.