Asymmetric Relaxations Through the Lens of Information Geometry
Alessandro Bravetti, Miguel Ángel García Ariza, Pablo Padilla
TL;DR
This work reframes relaxation toward thermodynamic equilibrium as a gradient flow on information-geometry, using the dual affine structure of equilibrium manifolds and the divergence $D^*_q=T_q D_{ ext{KL}}(ullet||q)$. It derives a general criterion for asymmetry in endoreversible relaxations via the Amari-Chentsov tensor $C^{ijk}$, and provides concrete analyses for classical isobaric and quantum ideal-gas relaxations, including a Mpemba-like non-isobaric scenario. The results yield a thermodynamic uncertainty relation and the Horse-Carrot theorem within this geometric framework, and reveal that asymmetry directions can reverse between classical and quantum regimes. Overall, the paper offers a unified geometric mechanism for understanding when cooling or warming dominates in a wide range of thermo-physical processes, with potential implications beyond thermodynamics to fields like machine learning and evolutionary dynamics.
Abstract
We frame Newton's Law of Cooling as a gradient flow within the context of information geometry. This connects it to a thermodynamic uncertainty relation and the Horse-Carrot Theorem, and reveals novel instances of asymmetric relaxations in endoreversible processes. We present a general criterion for predicting asymmetries using the Amari-Chentsov tensor, applicable to classical and quantum thermodynamics. Examples include faster cooling of quantum ideal gases and relaxations that resemble the Mpemba effect in classical ideal gases.
