their ability to predict unit-cell parameters is assessed. We start with an analysis of the unit-cell volumes. Overall, the agreement of cell volumes between the VASP MLPs, the MTPs, and DFT is excellent. This is evidenced by the relative deviations between force-field and DFT data plotted ...
Reference data (energy, force, and virial) for the structures were generated via DFT calculations using the VASP package (see “Methods” for details). The diversity of the initial training dataset ensured a robust initial NEP model that could be used to run MD and hybrid Monte Carlo and ...
The energy of each atom only depends on the atom type and position of the neighboring atoms within a relatively short radius. In other words, the interactions of an atom with its surroundings are assumed to be of a local nature. Again, this is similar to classical force fields, but in co...
The phonon eigenvalues computed from a force-constants matrix for a crystal structure describe the energy required to excite a specific vibrational mode within the crystal. For a potential to closely reproduce a phonon spectrum, it must accurately approximate thecurvatureof the potential energy surface ...
Cowen, B. J. & El-Genk, M. S. On force fields for molecular dynamics simulations of crystalline silica.Comput. Mater. Sci.107 Saika-Voivod, I., Sciortino, F., Grande, T. & Poole, P. H. Phase diagram of silica from computer simulation.Phys. Rev. E70, 061507 (2004). ...
We note our MSLP predicts a different order of stability of the magnetic states in FCC relative to VASP data, where a low spin FM state has slightly lower energy than the SL-AFM. On the other hand, DFT data from OpenMX predicts the same order of stability as our potential. This ...
(PBE) gradient-corrected exchange and correlation functional73and the projector augmented wave method74. During the active learning procedure, the VASP calculations were performed using Γ-centered Monkhorst-Packk-meshes where the number of divisions along each reciprocal lattice vector was chosen such ...
the error of the VASP MLPs is reduced even further by approximately two orders of magnitude. This is in stark contrast to the situation for UFF4MOF, where deviations amount to up to 14%. The errors of the machine learned force fields are somewhat larger for the anisotropic MOFs, but they...
Melting is the iconic first-order phase transition where a material changes from a solid to a liquid at the melting temperature,Tm. In equilibrium, this occurs when the Gibbs free energy, G=U−TS+PV, (1) of the solid equals that of the liquid. The energy absorbed atTm, the latent he...
Thus, here the force fields not only need to be For example, for MOF-5 the RMSD for the forces calculated with UFF4MOF amounts to 1.03 eVÅ−1, while it is only 0.02 eVÅ−1 for able to correctly predict '0 K' structures, but also need to capture the MTP and the VASP MLP...