The paper studies a type of random number-theoretic function called a Rademacher random multiplicative function. The idea is to assign to each prime number an independent random sign, either plus one or minus one with equal probability, and then extend this assignment to all positive integers by multiplication (so the value at a composite number is the product of the values at its prime factors). The central question is: if you add up these random values for all integers up to some large number x, how large does that sum typically get? Specifically, the paper pins down the size of fractional moments of this sum, meaning averages of the absolute value of the sum raised to a power between zero and two.
The key finding is a sharp phase transition. When the moment exponent q is below one half, the sum exhibits better-than-square-root cancellation, meaning it is smaller than you would expect from a simple random walk analogy. When q exceeds one half, the typical size of the sum grows like a power of the logarithm of x. Right at the critical value q equal to one half, the first absolute moment turns out to be of order the three-quarters power of the logarithm of the logarithm of x, a very slowly growing quantity. This kind of transition is reminiscent of phenomena seen in log-correlated random processes and in the mathematical theory of Gaussian multiplicative chaos, which describes random measures built from exponentials of log-correlated Gaussian fields.
The proof strategy connects the number-theoretic sum to integrals involving random Euler products, the infinite products over primes that naturally encode multiplicative structure. These objects are then analyzed by borrowing tools from Gaussian multiplicative chaos theory, exploiting the fact that the logarithm of such Euler products has a log-correlated structure similar to Gaussian fields. The hardest part of the argument is establishing a lower bound precisely at the critical moment q equal to one half, for which the authors also import ideas from the study of random walks, blending techniques from probability theory and analytic number theory.