Nonextensive statistical mechanics

Tsallis Entropy

A short note

tsallis.com · 2026

Abstract

As humans explore more complex systems and experience heavier tails, Tsallis entropy will attract much more attention.

1.  Background

The Tsallis entropy is a one-parameter generalization of the classical Boltzmann–Gibbs–Shannon entropy. Introduced by Constantino Tsallis in 1988 (and identical in form to the earlier Havrda–Charvát structural α-entropy), it underpins nonextensive statistical mechanics: a framework suited to systems with long-range correlations, multifractality, and power-law (heavy-tailed) statistics rather than the thin-tailed world of ordinary thermal equilibrium.

Sq = k (1 − ∑i piq) / (q − 1) (1)

For q → 1, Sq recovers the usual Shannon entropy. For q ≠ 1, the maximizers of Sq under suitable constraints are q-exponentials (including q-Gaussians), which exhibit heavier tails than the Gaussian and arise naturally in complex and non-equilibrium settings. A concise overview is available on Wikipedia: Tsallis entropy.

2.  Heavy tails and attention

Classical entropy and Gaussian-based inference remain the default toolkit of much applied statistics. As measurement and modeling push into domains where extremes dominate, such as financial returns, cascade processes, network traffic, extreme risk, and artificial intelligence and machine learning models, the mismatch between thin-tailed assumptions and observed sample paths becomes harder to ignore. Nonextensive entropy, and the associated q-distributions, offer a compact language for those regimes. Hence the claim of the abstract: wider encounter with complex systems and heavier tails will draw more serious attention to Tsallis entropy.

In the risk and fat-tails literature, Nassim Nicholas Taleb develops maximum-entropy constructions that include Tsallis-type distributions; see in particular the third edition of Statistical Consequences of Fat Tails, which adds material on maximum-entropy distributions including Tsallis [2].

References

  1. C. Tsallis. Possible generalization of Boltzmann–Gibbs statistics. Journal of Statistical Physics 52 (1988), 479–487. See also Tsallis entropy (Wikipedia).
  2. N. N. Taleb. Statistical Consequences of Fat Tails: Real World Preasymptotics, Epistemology, and Applications (third revised edition, 2025). Includes appendix material on maximum-entropy distributions (Tsallis). arXiv:2001.10488 · PDF.