Meaning of temperature in different thermostatistical ensembles

TL;DR

This paper explores the concept of temperature across different thermostatistical ensembles, clarifying the proper definitions and implications for thermodynamic laws, especially in non-equivalent ensembles and systems with discrete spectra.

Contribution

It clarifies the correct entropy definition in microcanonical ensembles and discusses the implications for temperature and thermodynamic laws, resolving common confusions.

Findings

Volume entropy satisfies thermodynamic laws exactly

Negative absolute temperatures are not physically achievable

Microcanonical ensemble definitions impact thermodynamic consistency

Abstract

Depending on the exact experimental conditions, the thermodynamic properties of physical systems can be related to one or more thermostatistical ensembles. Here, we survey the notion of thermodynamic temperature in different statistical ensembles, focusing in particular on subtleties that arise when ensembles become non-equivalent. The 'mother' of all ensembles, the microcanonical ensemble, uses entropy and internal energy (the most fundamental, dynamically conserved quantity) to derive temperature as a secondary thermodynamic variable. Over the past century, some confusion has been caused by the fact that several competing microcanonical entropy definitions are used in the literature, most commonly the volume and surface entropies introduced by Gibbs. It can be proved, however, that only the volume entropy satisfies exactly the traditional form of the laws of thermodynamics for a broad class of physical systems, including all standard classical Hamiltonian systems, regardless of their size. This mathematically rigorous fact implies that negative 'absolute' temperatures and Carnot efficiencies $>1$ are not achievable within a standard thermodynamical framework. As an important offspring of microcanonical thermostatistics, we shall briefly consider the canonical ensemble and comment on the validity of the Boltzmann weight factor. We conclude by addressing open mathematical problems that arise for systems with discrete energy spectrum.