The Rogers-Ramanujan continued fraction is a classical object in number theory, a specific infinite fraction whose values encode deep arithmetic information. This paper investigates two ways to measure this object through a technique called Mellin transforms, which are a kind of weighted averaging that converts a function into a new function of a complex variable. The authors focus on two flavors: ordinary moments, which weight the continued fraction by a power of the variable, and logarithmic moments, which additionally insert a logarithm. The key insight is that the continued fraction has two natural "boundary behaviors" at special points called cusps, and the interplay between these two boundary descriptions controls the analytic structure of both types of moments.
A central finding connects these moments to some of the most mysterious numbers in mathematics: the odd values of the Riemann zeta function, like zeta(3), zeta(5), and so on. Through formulas originally anticipated by Ramanujan, the logarithmic moments factor into products involving zeta functions and related objects called Dirichlet L-functions. The paper shows that translating between integer and half-integer input values gives an invertible coordinate change that mixes these odd zeta values together in a precise way. The authors also prove that certain numerical coefficients appearing in the cusp expansions are quadratic irrational numbers, meaning they satisfy simple degree-two polynomial equations with integer coefficients, and they find explicit minimal polynomials for all of them.
These structural results have concrete analytic consequences. The logarithmic moment generating function turns out to have a pole at every positive perfect square, and the authors obtain an explicit partial fraction expansion that converges absolutely. For ordinary moments, a two-cusp expansion involving incomplete gamma functions provides a meromorphic continuation to a wide region of the complex plane. Using a general technical tool they call a Mellin lemma, they conclude that these generating functions are not holonomic, meaning they cannot satisfy any linear differential equation with polynomial coefficients. The paper rounds out with symmetry identities, decay estimates across vertical strips in the complex plane, and quantitative bounds on truncation errors.