Fig. 2: Accuracy and efficiency of FermiNet-DMC on single atoms.
From: Towards the ground state of molecules via diffusion Monte Carlo on neural networks
a The blue line shows the energy of a 2-layer FermiNet as a function of training iteration for a Be atom. The orange symbols show the DMC energy obtained with the trial wavefunction at the 104th, 105th and 5 × 105th training iteration, respectively. The red dashed line shows the final convergence of VMC energy. b–d The three isosurfaces show the three-dimensional cuts of the full 11D nodal structure obtained at different training iterations (see Methods, section “Nodal structure and wavefunction visualization for plotting details”). e The calculated energies of atoms with respect to the reference ground state energy as the number of electrons increases with a 2-layer network. The dashed lines are linear fittings on the second and the third period elements, respectively. f The energy error with different settings of FermiNet for a Ne atom. Three variables are respectively the number of layers (L), the number of determinants (D) and the width (W) of each layer. All the energy reference values are from Chakravorty et al.57.
