The paper investigates a family of mathematical expressions built from cosines evaluated at special angles related to roots of unity, called cyclotomic cosine evaluations. A classical theme in this area is that when these expressions have enough symmetry, their values become highly constrained, eventually settling into simple, predictable patterns. The authors want to understand what happens when you move beyond this highly symmetric setting, and in particular, which features of the symmetric case survive and which do not.
The first main finding is essentially a negative result. The authors show that requiring the evaluated values to be rational numbers, which is a natural Galois-theoretic condition, places almost no real constraint on how complicated the underlying formulas can be. Specifically, any sequence of rational numbers you might want can be realized by a compatible family of polynomials whose complexity stays bounded. This means that rationality of values alone is far too weak a condition to force any kind of stable, simple long-term behavior in the formulas themselves.
The second and third findings are more constructive and positive. When the expressions are built from explicit patterns organized by multiplicative structure, the authors prove that the evaluated values do eventually collapse into simple linear functions of the level, with coefficients determined by a natural Fourier-like decomposition. Furthermore, breaking the index set into orbits according to greatest common divisors produces an exact formula involving classical number-theoretic objects called Ramanujan sums, which measure multiplicative interference patterns between integers. Special cases like semiprime and prime-power levels fall out cleanly from this general formula. Taken together, the results draw a precise boundary between where symmetry-based rigidity genuinely breaks down and where a different, more combinatorial kind of structure takes over to produce equally clean behavior.