Fast free-energy estimation for complex systems with nonreversible simulated tempering

We discuss free-energy estimation for complex systems using simulated tempering with Wang--Landau-type feedback. We show that introducing nonreversible dynamics in temperature space and multiple temperature updates per step significantly accelerates the convergence of the free-energy estimate. We then apply the algorithm to the three-dimensional Edwards-Anderson spin-glass model. Our results show that the algorithm can precisely estimate the free energy up to a large system size $L \simeq 64$, and that subsequent equilibrium simulated tempering simulations with the estimated free energy can rapidly sample from the Boltzmann distributions over a wide temperature range that covers the spin-glass transition temperature. The convergence time of our algorithm grows much more slowly with system size than that of another known algorithm for on-the-fly free-energy estimation.