Machine Learning for Three-Body Recombination Rates Across 8 Decades of Energy

Three-body recombination reactions are ubiquitous in atomic and molecular physics and are essential for understanding a wide variety of phenomena from losses in traps to the microscopic underpinnings of nucleation. The 3-body problem of solving the classical dynamics of individual reactants, which is needed for ab initio studies, involves chaos and as such must be addressed computationally, leading to high computational costs for determining reaction rates. We present a simple feed-forward neural network approach for predicting the opacity function of atom-atom-ion and atom-atom-atom three-body recombination reactions using only a small training dataset. The models developed are succesful over a wide range of energies, indicating the model learned important underlying physics. Our approach thus offers an avenue for circumventing additional computation of the three-body dynamics, making predictions that are much faster than the corresponding simulations while maintaining an accuracy comparable to the ab initio method.