Custom code developments

1. Classical potential learning from reference calculations

- Finite temperature, large-scale phenomena such as charge density wave or domain formations play important roles in materials properties but first-principles calculation for them requires prohibited expensive calculation cost that constructing accurate classical potential are highly desired. In my approach, interatomic potential is represented with symmetrized, pairwise and polynomial functions, and the coefficients and number of basis functions are fitted or machine-learned to reference calculations. Molecular dynamics simulations or coordinate optimization can be performed for the obtained potential, and two applications (incommensurate charge density wave of monolayer 2H-NbSe2 and 2x2 CDW domains and stacking structures of Kagome metal CsV3Sb5) are shown in Fig. (e) and (f). For latter preliminary calculations, 256 cores are used for 150x150x4 supercell. Due to the almost linearity of parallelization method, high-performance computing resource can treat stacking ordering procedure in a larger supercell.       

2. Phonon-Postprocessing codes: Lattice Wannier function, phonon unfolding

- Materials with a structural phase transition or charge density wave can be analyzed with soft (unstable) phonon mode. The analysis becomes visually appealing and gives insight when the concept of lattice Wannier function (optimized local mode) and unfolded phonon band are applied.

3. Finite bias STM simulation code

- Finite bias of STM tip polarizes the sample and simultaneously, the sample screens the polarization. This effect is usually ignored in the comparison between experiment and conventional STM simulations included in DFT codes, but it is not at all small for molecules or atomically thin 2D materials. From my first-principles finite bias STM simulations, monolayer TCNQ layer, which becomes metallic due to the charge transfer from Au substrate, reduces applied bias by 10% as shown in Fig. (b).  

4. Random diffusion simulation code

- The fast oxygen diffusion is core properties for good ceramic solid oxide fuel cell. In many cases, the substitutional cation dopant tends to increase the oxygen diffusion rate, but the reason is not clear. From DFT calculations, energy barriers of various oxygen diffusion path can be calculated, and they are combined with diffusion simulations code, which can reveal the effect of randomly distributed cation dopants. For converged results, 500,000-1,000,000 particles with 500-1000 random walking steps are required. This process is independent that high-performance computing resource can simulate concentration-dependence of various solid oxide fuel cell materials in a few days.

5. Field emission simulation code

- Overcoming the degradation of carbon nanotube field emitter, more stable boron nitride (BN) nanotube is suggested by us theoretically. The performance and improvement principles are also suggested from real-time field-emission simulation of BN nanotube.