A large deviation approach to superstatistics: thermodynamic duality symmetry between conjugate variables

Abstract: Superstatistics generalizes Boltzmann statistics by assuming spatio-temporal fluctuations of the intensive variables. It has many applications in the analysis of experimental and simulated data. The fluctuation of the intensity variable is the key to the validity of superstatistical theory, but the law of its distribution is still unclear. In the framework of large deviation theory, we show that the fluctuation of the intensive variable of superstatistics emerges naturally from measurements in the large data limit. Combining Bayes' theorem, we demonstrate the conditional probability distribution of the intensity variable also follows the Boltzmann statistics and the conjugate variable of the intensive variable is the extensive variable, indicating a thermodynamic duality symmetry between conjugate variables in the superstatistical systems. A new thermodynamic relation between the entropy functions of conjugate variables is obtained. We utilized a simple Ising model with fluctuating temperature to verify the dual relationship between temperature and energy. Our work may contribute to the understanding of statistical physics in complex systems and Bayesian inference.