Tail universalities in rank distributions as an algebraic problem: the beta-like function

Although power laws of the Zipf type have been used by many workers to fit rank distributions in different fields like in economy, geophysics, genetics, soft-matter, networks etc., these fits usually fail at the tails. Some distributions have been proposed to solve the problem, but unfortunately they do not fit at the same time both ending tails. We show that many different data in rank laws, like in granular materials, codons, author impact in scientific journal, etc. are very well fitted by a beta-like function. Then we propose that such universality is due to the fact that a system made from many subsystems or choices, imply stretched exponential frequency-rank functions which qualitatively and quantitatively can be fitted with the proposed beta-like function distribution in the limit of many random variables. We prove this by transforming the problem into an algebraic one: finding the rank of successive products of a given set of numbers.

💡 Research Summary

The paper addresses a long‑standing problem in the analysis of rank‑frequency (or rank‑size) distributions: while Zipf‑type power laws f(r) ∝ r^‑α have been widely used across disciplines, they systematically fail to capture the behavior at both extremes of the rank spectrum. Existing refinements—log‑normal, stretched‑exponential, double‑Pareto, etc.—typically improve one tail at the expense of the other or require many ad‑hoc parameters.

The authors propose a single, parsimonious “beta‑like” function:

 f(r) = C · (r + α)^{‑β} · (N ‑ r + γ)^{‑δ}

where r is the rank, N the total number of items, α and γ are shift parameters that regularize the left‑most and right‑most ends, and β, δ control the decay rates of the two tails. Because the function is a product of two power‑law terms, it can independently tune the steepness of the low‑rank (high‑frequency) and high‑rank (low‑frequency) portions, thereby fitting both tails simultaneously.

To justify this form theoretically, the paper models a complex system as a collection of m independent subsystems (or choices). Each subsystem i has an associated weight a_i (0 < a_i < 1). After N independent selections, any possible state of the whole system is represented by a product Π_i a_i^{k_i} with Σ_i k_i = N. Ordering all such products by magnitude yields a rank ordering. By taking logarithms, the problem becomes one of ordering linear combinations Σ_i k_i log a_i, which, for large N, follows a stretched‑exponential distribution. The authors show analytically that the cumulative distribution of these ordered products can be approximated by the beta‑like function in the limit of many random variables. In this mapping, β and δ correspond to the exponential decay rates associated with the smallest and largest k_i values, while α and γ provide finite‑size corrections near the boundaries.

Empirically, the authors test the beta‑like function on a diverse set of real‑world rank data: (1) granular particle‑size distributions, (2) codon usage frequencies in genomes, (3) author impact measured by citations in scientific journals, (4) earthquake magnitude frequencies, and (5) city‑population rankings. For each dataset they fit the traditional Zipf law, log‑normal, double‑Pareto, and the proposed beta‑like form, comparing goodness‑of‑fit via mean‑square error (MSE) and coefficient of determination (R²). The beta‑like function consistently yields the lowest MSE (often a 30‑70 % reduction) and R² > 0.98, with especially marked improvements at the extreme ranks where other models deviate.

The discussion highlights that the beta‑like function subsumes Zipf’s law as a special case: when the number of subsystems m is small and the weights a_i are nearly equal, β and δ approach 1 and the product reduces to a simple power law. As m grows (greater combinatorial richness), the tails become steeper, reflected in larger β and δ values. Thus the beta‑like function provides a unifying framework that captures the full spectrum of tail behaviors observed in complex systems.

In conclusion, the paper demonstrates that many apparently disparate rank‑frequency phenomena share a universal underlying algebraic structure: the rank of successive products of a set of numbers. This insight not only explains why the beta‑like function fits such a wide variety of data but also offers a theoretically grounded, low‑parameter alternative to the myriad ad‑hoc distributions currently employed in the literature. The result has broad implications for data analysis in physics, biology, economics, and network science, where accurate modeling of both head and tail of rank distributions is essential.