Power laws, Pareto distributions and Zipf’s law

Abstract

Power-law distributions characterize a diverse array of natural and anthropogenic phenomena, ranging from word frequencies and city populations to earthquake magnitudes and scientific citations. These distributions are defined by an inverse relationship between the probability of a value and a power of that value, resulting in heavy-tailed behavior and the absence of a characteristic scale. Accurate empirical identification of power laws requires rigorous statistical frameworks, particularly maximum likelihood estimation and the analysis of cumulative distribution functions, to mitigate the biases associated with least-squares fitting on logarithmic scales. Mathematically, these functions are unique in their scale invariance, though they often present divergent moments that render traditional measures like the mean or variance ill-defined. The emergence of power-law behavior can be attributed to several distinct stochastic and physical mechanisms. Preferential attachment processes, such as the Yule process, generate power laws through growth proportional to current size. In physics, critical phenomena near continuous phase transitions produce scale-free behavior as characteristic lengths diverge, while self-organized criticality allows systems to dynamically maintain a critical state. Additional explanatory models include random walks, multiplicative processes, and highly optimized tolerance. Understanding these distributions is essential for characterizing the structural and dynamical properties of complex systems across the physical and social sciences. – AI-generated abstract.