The paper studies a specific class of functions over finite fields, called "read-once polynomials." These are formulas built using addition, multiplication, and constants, where each input variable appears at most once. Think of them as arithmetic expressions like (x + y) times z, where no variable is reused. When you feed such a function random, independently chosen inputs from a finite field, you get a probability distribution over the field's elements, describing how often each output value appears. The central question is: what distributions are actually achievable this way?
The main result identifies a precise geometric region, called a body, that contains every distribution produced by any read-once polynomial over a finite field with k elements (for k at least 4). The condition defining this region involves only two numbers from the sorted distribution: the second-largest probability and the smallest probability. Intuitively, the constraint says the output cannot be too "lopsided" in a particular way. The key technical insight is that this region is stable under the operations of the field, meaning if two distributions both satisfy the condition, then combining them through either field addition or field multiplication produces a distribution that also satisfies it. This stability holds more broadly for any "quasigroup" operation, a generalization of a group where division is always possible.
The paper also works out the volume of this region in the space of probability distributions, finding an exact formula and showing that the volume shrinks exponentially as k grows, with a growth rate tied to the specific constant 0.2183... Notably, every result in the paper, including both stability theorems, the volume formula, and its exponential rate, has been formally verified using Lean 4, a computer-based proof assistant. This means the mathematical arguments have been checked with machine-level rigor, providing unusually strong confidence in the correctness of the findings.