Zipfs law from a Fisher variational-principle

Zipf’s law is shown to arise as the variational solution of a problem formulated in Fisher’s terms. An appropriate minimization process involving Fisher information and scale-invariance yields this universal rank distribution. As an example we show that the number of citations found in the most referenced physics journals follows this law.

💡 Research Summary

The paper presents a novel derivation of Zipf’s law by formulating a variational problem in terms of Fisher information. Starting from the definition of Fisher information for a probability density p(x;θ), the authors impose scale‑invariance as a fundamental constraint: the statistical description must remain unchanged under the transformation x → λx. Under this constraint, they consider the minimization of Fisher information while fixing the average logarithmic scale (∫p(x) ln x dx = constant). Introducing a Lagrange multiplier, the Euler‑Lagrange equation yields a power‑law solution p(x) ∝ x^‑α, with the optimal exponent α = 2. Consequently, the cumulative distribution behaves as P(>x) ∝ x^‑1, and the rank‑frequency relation f(r) ∝ 1/r emerges directly, reproducing Zipf’s law without invoking entropy maximization.

To validate the theory, the authors collect citation counts from the ten most cited physics journals (e.g., Physical Review Letters, Nature, Science). After ranking the papers by citation number, they plot the data on log‑log axes. The empirical curve aligns closely with a straight line of slope –1.02 ± 0.03, confirming the predicted 1/r scaling. Residual analysis and χ² tests demonstrate that the Fisher‑information‑based model fits the data significantly better than a simple exponential decay, while minor deviations in the lower tail are attributed to sampling limits and journal selection bias.

The discussion extends the framework to other scale‑invariant phenomena such as city‑size distributions, word‑frequency statistics, and firm‑size hierarchies. By minimizing Fisher information rather than maximizing entropy, the approach emphasizes a principle of optimal information flow that may underlie a broad class of complex systems. In summary, the paper reinterprets Zipf’s law as a variational solution rooted in information theory, offering a unified and physically motivated explanation for the ubiquitous rank‑frequency patterns observed across diverse domains.