Neural network variational Monte Carlo for positronic chemistry

Gamma rays produced by positron annihilation are used as a sensitive probe of matter at atomic length scales. Technologies for manipulating positrons and studying their interactions with ordinary matter are rapidly progressing. This motivates the development of accurate ab initio methods for modelling positronic interactions with molecular matter. Here, we apply the recently developed Fermionic neural network (FermiNet) wavefunction ansatz to the problem of finding the ground-state properties of mixed positron-electron systems. We find that FermiNet produces highly accurate, in some cases state-of-the-art, ground-state energies across a range of atoms and small molecules with a wide range of qualitatively distinct positron binding characteristics. We highlight the capabilities of our method by calculating the positron binding energy of the challenging non-polar Benzene molecule.