A structural optimization algorithm with stochastic forces and stresses

We propose an algorithm for optimizations in which the gradients contain stochastic noise. This arises, example, structural when computations of forces and stresses rely on methods involving Monte Carlo sampling, such as quantum or neural network states, are performed devices that have intrinsic Our proposed is based combination two ingredients: update rule derived from steepest-descent method, a staged scheduling targeted statistical error step size, with position averaging. compare it commonly applied algorithms, including some latest machine learning optimization methods, show consistently performs efficiently robustly under realistic conditions. Applying this algorithm, we achieve full-degree solids using ab initio many-body computations, by auxiliary-field plane waves pseudopotentials. A potential metastable structure Si discovered density-functional calculations synthetic noisy forces. Energy accurate post-density-functional-theory often involve uncertainties. study proposes originating methods.