Optimal Estimation of Temperature

Shaoyong Zhang, Zhaoyu Fei, Xiaoguang Wang

Published 2025-01-07Version 1

Over the past century, the Boltzmann entropy has been widely accepted as the standard definition of entropy for an isolated system. However, it coexists with controversial alternatives, such as the Gibbs entropy. These definitions, including the Boltzmann entropy, exhibit certain inconsistencies, both mathematically and thermodynamically. To address this challenge, we introduce the estimation theory in statistical inference into the study of thermodynamics and statistical physics for finite-sized systems. By regarding the finite-sized system as a thermometer used to measure the temperature of the heat reservoir, we show that optimal estimation of temperature yields the corresponding entropy formula for an isolated system. In the single-sample case, optimal estimation of inverse temperature (or temperature) corresponds to the Boltzmann entropy (or Gibbs entropy). These different definitions of entropy, rather than being contradictory, apply to optimal estimation of different parameters. Furthermore, via the Laplace transform, we identify a complementarity between estimation of temperature and system's energy, a concept suggested by Niels Bohr. We also correct the energy-temperature uncertainty relation, as expressed by the Cram\'{e}r-Rao bound, in the large-$N$ limit. In the multiple-sample case, we generalize the definitions of both Boltzmann entropy and Gibbs entropy to achieve optimal estimation of temperature, revealing the tight connection between statistical inference and Terrell Hill's nanothermodynamics.