A classification of nonequilibrium steady states based on temperature correlations

Although generalized ensembles have now been in use in statistical mechanics for decades, including frameworks such as Tsallis' nonextensive statistics and superstatistics, a classification of these generalized ensembles outlining the boundaries of validity of different families of models, is still lacking. In this work, such a classification is proposed in terms of supercanonical and subcanonical ensembles, according to a newly defined parameter, the inverse temperature covariance parameter $\mathcal{U}$. This parameter is non-negative in superstatistics (and is equal to the variance of the inverse temperature) but can be negative for other families of statistical ensembles, adquiring then a broader meaning. It is shown that $\mathcal{U}$ is equal for every region of a composite system in a steady state, and examples are given of supercanonical and subcanonical states.