The paper investigates how certain mathematical structures called "amenable subrelations" fit inside larger structures called "treeable equivalence relations." To understand the setting roughly: equivalence relations partition a mathematical space into classes of mutually related points, and "treeable" means these classes can be organized using tree-like graphs. "Amenable" is a technical property reflecting a kind of internal averaging or regularity. The authors use a geometric tool they developed in a companion paper, which tracks how the Radon-Nikodym derivative (a measure-theoretic object capturing how one measure distorts another) behaves across the structure.
A central result concerns something called "ends," which are the distinct ways of escaping to infinity along a tree. The authors show that a theorem by Scott Adams about selecting ends consistently across equivalence classes has a precise geometric interpretation: the ends Adams selects for a given class are exactly those that remain significant, or "nonvanishing," along that class in terms of the Radon-Nikodym topography. Building on this, they prove that every well-behaved amenable subrelation is contained in an essentially unique largest amenable subrelation, a result that mirrors how one might expect maximal objects to behave in algebra or geometry but requires substantial new work in this setting.
The practical payoff comes in applications to group theory. A group is called "treeable" if its natural action on a probability space can be organized by a treeable equivalence relation, a property with deep connections to how groups can be decomposed and studied. Using their structural results as a criterion for ruling out treeability, the authors prove two notable theorems: any nonamenable locally compact group whose "amenable radical" (its largest normal amenable subgroup) is noncompact cannot be treeable, and any nonamenable treeable group that lacks a symmetry property called unimodularity must contain a compact open subgroup. These results significantly constrain which groups can be treeable and answer open questions in measured group theory.