We plan to prepare and perform large scale simulation on K computer to establish new standards of materials science. In this paper, we introduce specific examples in the field of materials science; phase diagram calculation and high accuracy simulation on hydrogen storage materials. Our own ab initio calculation code, TOMBO is “state of readiness” to be distributed and computational materials science in “petaflop era” using it is discussed.
First principles molecular dynamics (FPMD) simulations for electrochemical systems are reviewed. Effective screening medium (ESM) is utilized to simulate the Pt/water interface with a bias. H-down structure of water molecules and electron transfer reaction on negatively charged Pt surface are simulated by FPMD. Polymer electrolyte membrane of Nafion and carbon hydride are simulated by the large scale FPMD with homogeneous electric field. Electroosmosis is observed. Fast ionic conductor, LiBH4, is simulated by order-N FPMD of 1200 atoms. Li ion diffusion path and its barrier are obtained.
Dynamical behaviors of nanoscale electric transport are discussed with presenting recent simulation results obtained by our group on the following three topics. First, the conductance fluctuation of single molecular bridges between metal electrodes due to temperature and surrounding water molecules are examined. Second, our attempt toward the understanding on the switching mechanism of solid electrolyte atomic switches is described. Finally, the ac transport in metallic carbon nanotubes is discussed.
Molecular dynamics studies for nano-scale biomaterials driven by the Next-Generation Supercomputer Project are roughly reviewed. In particular, calculations for spherical micelles, lipid bilayers, and viruses are presented. We discuss about the stability of SDS micelle as well as the solubilaization of immiscible solutes by it, the properties of real cell membranes, i.e. normal and cancer cells, and the future calculation plan for virus capsid. Development of the highly parallelized general-purpose molecular dynamics simulation software(Modylas) is also introduced.
The structures of Nb-doped anatase TiO2 (TNO) were calculated using density functional theory (DFT)-based first-principle method. In order to clarify the role of oxygen vacancies, periodic unit cells with several combinations of dopant atoms and oxygen vacancies were investigated. The same calculation scheme was adapted to W-doped anatase TiO2 (TWO), and Nb-doped rutile TiO2 for a comparison. The results showed that the possibility of threebody complex in TNO is rather small, compared to the case of TWO and Nb-doped rutile TiO2. In the latter cases, a strong energy stabilization was observed in a linear W-VO-W and a bent VO-Nb-VO structure, respectively.
All-electron first-principles Green's function methods such as GW method, GW+Bethe-Salpeter method, and GW+T-matrix method were developed for massive parallel calculations using hundreds of thousands CPU cores. The benchmark test executed by using up to 1,024 CPU cores shows that the performances are scalable against the number of CPU cores. To ascertain the accuracy of present methods, in addition, we applied these methods in calculating the single electron excitation energy spectra, the optical absorption spectra, and the Auger spectra of some small sized systems.
Because of the undesired computational scaling, straightforward quantum chemical treatments of nano-systems are hopeless even if the next-generation supercomputer “K” is released. Therefore, in the field of quantum chemistry, development of linear-scaling computational methods has been one of the central issues in the last two decades. In this paper, we briefly summarize our recent developments in the divide-and-conquer (DC) quantum chemical method, which is one of the linear-scaling methods. Although the original DC method by Yang was formulated for the Hartree-Fock (HF) and Kohn-Sham density functional theories, it has been extended to the accurate post-HF correlation theories by virtue of the energy density analysis. We have shown that the DC methods achieve linear-scaling computational time without reducing the accuracy.