The paper studies a type of random sum that combines two ingredients: a "random multiplicative function" and a "periodic weight." The random multiplicative function assigns a random complex number to each integer in a way that respects multiplication (so the value at a product of two numbers equals the product of the values at those numbers). The periodic weight is a smoothly varying pattern that repeats with a period tied to an irrational number, meaning the pattern never exactly repeats on the integers and spreads out in a mathematically rich way. The central question is: when you add up N terms of this combined object, does the resulting sum behave like a standard bell curve (more precisely, a complex Gaussian) as N grows large?
The authors find a precise condition that determines whether or not this Gaussian behavior holds. The condition depends on how well the irrational number can be approximated by fractions, which is a classical concern in number theory. Some irrational numbers are "too well approximated" by rationals, and in those cases the sum can deviate from Gaussian behavior. This is a sharp result in the sense that the condition is both necessary and sufficient, meaning it fully characterizes when the central limit theorem holds and when it fails.
The paper also proves a cleaner, unconditional result when the sum is restricted to integers whose prime factors are all larger than some growing threshold. This restriction removes the small primes, which are responsible for the correlations that can disrupt Gaussian behavior. Under this restriction, a central limit theorem always holds, no matter how slowly the threshold grows and no matter which irrational number is used. This suggests that the potential failure of Gaussian behavior in the general case is entirely driven by the contribution of small primes interacting with the periodic weights.