The Hidden Bridge

Population annealing is a practical sampling algorithm: take a population of particles at one temperature, slowly change the temperature, and at each step reweight and resample — kill particles with low weight, clone particles with high weight. It is treated as a heuristic, an engineering compromise between exactness and computational cost.

Ohzeki (arXiv:2603.16056) proves that the reweighting step is not a heuristic at all. It is the exact analytical solution to the Schrödinger bridge problem — the optimal way to transport a probability distribution from one state to another while minimizing a path-space divergence. The “approximate” algorithm was solving an optimal transport problem all along.

The connection goes deeper. The thermodynamic work — the energy cost of driving the system out of equilibrium, which practitioners try to minimize as a practical matter — turns out to be the optimal control potential in the variational formulation. The quantity that thermodynamics treats as waste (non-equilibrium work) is the quantity that mathematics treats as the solution (optimal control). The Jarzynski equality, which relates non-equilibrium work to equilibrium free energy differences, is reinterpreted as a consistency condition within the Donsker-Varadhan variational principle.

The algorithm was not an approximation to the theory. The algorithm was the theory, discovered empirically before its mathematical identity was recognized. The resampling step — which looks like a crude intervention (kill the bad ones, copy the good ones) — is the instantaneous projection that solves the bridge problem without the iterative schemes that the optimal transport literature requires.

The heuristic is the exact solution. The practical workaround is the mathematical optimum. The thing that looked like engineering was always mathematics, waiting to be recognized.