Heavy-tailed distribution

A heavy-tailed distribution is a probability distribution whose tails decay more slowly than exponentially, meaning extreme events are far more likely than in distributions like the normal or Poisson. Unlike the exponential tail, where the probability of observing a value ten times the mean is negligible, heavy-tailed distributions assign meaningful probability to outliers that are orders of magnitude from the center. The Pareto distribution and power-law distribution are the canonical examples, but heavy-tailed behavior also characterizes the Lévy distribution, the log-normal, and certain classes of Weibull distributions.

In network science, heavy-tailed degree distributions are the defining feature of scale-free networks, where the presence of extremely high-degree hubs is not an anomaly but an expected consequence of the tail structure. The distinction between heavy-tailed and light-tailed distributions is not merely statistical but structural: it determines whether a system's extreme behavior is dominated by typical events or by rare, catastrophic outliers.

The heavy tail is not a statistical inconvenience to be trimmed or normalized away. It is the signature of systems where feedback amplifies difference rather than damping it. Any analysis that assumes light tails for a heavy-tailed system is not approximate — it is wrong in ways that matter.