Dielectric Constant of Liquid Water Determined with Neural Network Quantum Molecular Dynamics

Self-Consistent Determination of Long-Range Electrostatics in Neural Network Potentials

“…Here, the effective external potential refers to the Kohn-Sham effective potential, which includes the external potential and the self-consistently determined long-range electric fields. Therefore, the NEM suggests that the electronic density, and consequently the positions of the MLWFCs, are 'nearsighted' with respect to the effective potential, but not to the atomic coordinates, contrary to what has been assumed in previous work that also uses local geometric information of atoms as input to neural networks 45,46 . An atom located at r will affect the effective potential at r, even if r is far from r, through long-range electrostatic interactions.…”

“…An atom located at r will affect the effective potential at r, even if r is far from r, through long-range electrostatic interactions. Consequently, current approaches to generating NN models can only predict the position of MLWFCs for a purely short-range system without longrange electrostatics, such as the GT system 45,46 . We exploit this fact and use established NNs to predict the locations of the MLWFCs in the GT system 45 .…”

“…Consequently, current approaches to generating NN models can only predict the position of MLWFCs for a purely short-range system without longrange electrostatics, such as the GT system 45,46 . We exploit this fact and use established NNs to predict the locations of the MLWFCs in the GT system 45 . To do so, we create a local reference frame around each water molecule (Fig.…”

Dielectric dispersion characteristics of the phospholipid bilayer with subnanometer resolution from terahertz to mid-infrared

“…As the phospholipid bilayer and the water are respectively the major constituents of the biomembrane and the intracellular fluid, the optical/dielectric properties of the phospholipid bilayer and the water are thus reasonable indicators of those of the myeline sheath and the axon, respectively. The optical/dielectric constants of the water over a wide frequency range from direct current to ultraviolet at various temperature have been thoroughly investigated, and plenty of theoretical and experimental data can be acquired from the available literature ( Hale and Querry, 1973 ; Segelstein 1981 ; Buchner et al, 1999 ; Praprotnik and Janežič, 2005 ; Heyden et al, 2010 ; Midi et al, 2014 ; Rowe et al, 2020 ; Krishnamoorthy et al, 2021 ). The studies on the biomembrane or phospholipid bilayer also have been started for long.…”

Towards computational polar-topotronics: Multiscale neural-network quantum molecular dynamics simulations of polar vortex states in SrTiO3/PbTiO3 nanowires

“…The primary challenge is to computationally describe their complex polarization dynamics extending many spatiotemporal scales as they emerge, while retaining quantum-mechanical accuracy. This challenge could be met by recent successes in machine learning based interatomic potentials to perform neural-network quantum molecular dynamics (NNQMD) simulations with ab initio accuracy but at a fraction of computational cost (Behler, 2015;Krishnamoorthy et al, 2021). In fact, we have successfully studied optical control of polar topological structures in PTO using NNQMD simulations on massively parallel supercomputers, which were trained by excited-state quantum molecular dynamics (QMD) simulations (Linker et al, 2022).…”