The paper addresses a deceptively simple question about physical systems: when you take a system out of equilibrium and let it interact with a heat bath at a fixed temperature, does it eventually settle into a true equilibrium state with the lowest possible free energy? Free energy is a thermodynamic quantity that measures how much useful work a system can do, and it is supposed to decrease over time as a system relaxes toward equilibrium. While physicists and students take this for granted, proving it rigorously in mathematics turns out to be surprisingly tricky, because free energy has a technical mathematical property called lower semicontinuity that makes the limiting behavior hard to pin down precisely.
The specific setting studied here is the classical Ising model with Glauber dynamics, a well-known mathematical framework for describing how magnetic spins on a lattice randomly flip over time under the influence of temperature. Mathematicians have long known that free energy decreases during this process, but whether it actually converges all the way down to the equilibrium value had only been proven in regimes where the system has a unique equilibrium state. The harder and more interesting case is when the temperature is low enough that multiple equilibrium states can coexist, for example when a magnet can be magnetized in either direction. In this phase-coexistence regime, existing mathematical tools broke down.
The authors resolve this open problem by introducing a new mathematical inequality called a coarse modified log-Sobolev inequality. This inequality provides a quantitative way of measuring how quickly free energy is being dissipated, and crucially it holds at every positive temperature, including in the difficult phase-coexistence regime. By establishing this inequality, they prove that the free energy of the evolving system does indeed converge to the free energy of an equilibrium state in full generality, giving rigorous mathematical backing to something that physicists and students have always assumed to be true.