| Details: |
The projected augmented plane-wave (PAW) method has been proven instrumental for ab initio calculations, especially in the studies of solid. In a couple of recent articles [1,2], we have shown that evolutionary tuning some of the PAW parameters can provide a productive and consistent way to improve the efficacy and accuracy of the PAW datasets substantially. The bio-inspired stochastic optimization strategy [2] we have developed and implemented has been named as "Evolutionary Generator of projector augmented wave datasets" (EPAW-1.0). EPAW-1.0 is an automated hybrid recipe that integrates evolutionary computing with density functional theory calculations. The self-learning evolutionary algorithms in EPAW-1.0 adaptively tune some of the PAW parameters (such as different radii, reference energies and shape functions) to match the equation of states (EoS) produced by standard all-electron method while maintaining all the realizations from PAW method. The program imposes various constraints on logarithmic derivatives and basis sets features to avoid numerical instability, ghost states, and maximize transferability. The generated data-set produces a closest possible equation of state (EoS) with respect to a given target EoS for desirable pressure ranges. The high-pressure end allows us to probe compressed volumes as small as V0/2 and often smaller. Across the periodic table, the evolutionary optimized PAW (EPAW) datasets outperform all other standard PAW dataset libraries, ultra-soft, and norm-conserving pseudopotentials explored and mentioned in the celebrated endeavor by Lejaeghere et al. [3]
1) Sarkar K.*, Topsakal M., Holzwarth N.A.W., Wentzcovitch R.M. Journal of Computational Physics,347 (2017), pp.39-55
2) Sarkar K.*, Holzwarth N.A.W., Wentzcovitch R.M. Computer Physics Communications, (2018), 233 (2018), pp 110-122
3) K. Lejaeghere et al., Science (80-.), 351 (2016) aad3000-aad3000
|