Toward the Exact Exchange–Correlation Potential: A Three-Dimensional Convolutional Neural Network Construct
Code: N/A Model: 3D CNN Translation invariance: regularization term (loss) URL: http://yangtze.hku.hk/home/pub/179.pdf Year: 2019
In principle, the determination of the potential at any ri, vxc(ri) requires the knowledge of global electron density function ρ(r). Indeed, Nagai et al.32 [Neural-network-Kohn-Sham-exchange-correlation-potetial] proposed a neural network that maps ρ(r) to vxc(r) and as a proof of concept determined the vxc(r) of a one-dimensional two-particle spinless Fermion model via a neural network that one-to-one maps its ρ(r) to its vxc(r). Just as other existing machine-learnt xc functionals,26,27 their approach requires the knowledge of the global electron density function ρ(r) to determine the vxc(r) 32 and thus cannot be applicable for large molecules as their electron density functions are not accurately known.