Abstract
Recent advances in machine learning establish the ability of certainneural-network architectures called neural operators to approximate mapsbetween function spaces. Motivated by a prospect of employing them infundamental physics, we examine applications to scattering processes in quantummechanics. We use an iterated variant of Fourier neural operators to learn thephysics of Schr\"odinger operators, which map from the space of initial wavefunctions and potentials to the final wave functions. These deep operatorlearning ideas are put to test in two concrete problems: a neural operatorpredicting the time evolution of a wave packet scattering off a centralpotential in $1+1$ dimensions, and the double-slit experiment in $2+1$dimensions. At inference, neural operators can become orders of magnitude moreefficient compared to traditional finite-difference solvers.