Shannon entropy is real-valued. It measures the uncertainty of a probability distribution — the expected information content of an observation. The logarithm of a probability is always real and negative, and the expectation is always real and non-negative. There is no imaginary component. Entropy lives on the real line.
Until you extend probability measures into the complex plane.
Complex-valued probability measures assign complex numbers to events. This sounds pathological — probabilities should be real, non-negative, summing to one. But the construction is mathematically rigorous: the complex measure is a phase-modulated extension of a classical distribution, where the modulus retains the classical probability and the phase introduces a new degree of freedom.
Note (2026-11-10): I checked the abstract and searched the full text of arXiv 2603.12297. The paper defines complex probability measures as phase-modulated extensions of classical ones, proposes complex entropy, complex divergence and a complex metric, compares them with Shannon entropy and Kullback-Leibler divergence, and says it elucidates a profound formal analogy with Feynman's path integral, suggesting a conceptual bridge; for two-sample testing it gives a procedure with advantages and limitations and a conceptual simulation. The essay says the correspondence with the path integral is not an analogy and is a structural isomorphism, and says the framework enables the test; the paper says formal analogy and presents the testing application conceptually.
From this construction, three new information-theoretic quantities emerge: complex entropy, complex divergence, and complex metric. Complex entropy measures distribution uniformity through phase coherence rather than amplitude alone. Complex divergence provides an asymmetric dissimilarity that captures structural differences invisible to classical divergence measures. The complex metric is a true distance — symmetric, satisfying the triangle inequality.
The formal structure of complex entropy turns out to parallel Feynman's path integral formulation of quantum mechanics. The integral over complex-valued probability measures mirrors the integral over complex amplitudes in the path integral. This is not an analogy chosen for rhetorical effect — it is a structural isomorphism between the entropy integral and the quantum propagator.
The practical grounding prevents this from being pure abstraction. The framework enables nonparametric two-sample hypothesis testing — distinguishing whether two datasets come from the same distribution — using the complex metric. The phase information captures distributional differences that amplitude-only measures miss.
Information theory and quantum mechanics share a mathematical root that neither field constructed deliberately. The complex entropy exposes it.