Project For Advancement of Software Usability in Materials Science

TL;DR

PASUMS addresses the fragmentation and usability barriers in computational materials science by developing and distributing open-source software across first-principles, lattice-model, and data-driven workflows. The project emphasizes interoperability, documentation, and scalable deployment on ISSP supercomputers, delivering tools such as OpenMX, ESM-RISM, RESPACK, H-wave, DCore, HΦ, TeNeS, abICS, PHYSBO, and supporting infrastructure like MateriApps Installer and HTP-Tools. Key contributions include enabling downfolding to effective models, DMFT and tensor-network methods, Bayesian optimization for materials design, and high-throughput data generation, all integrated through community-oriented licensing and documentation. The work accelerates materials discovery by providing a coherent software ecosystem and data-generation capabilities, with future goals of integrated software frameworks and data repositories to facilitate materials informatics at scale.

Abstract

The Institute for Solid State Physics (ISSP) at The University of Tokyo has been carrying out a software development project named ``the Project for Advancement of Software Usability in Materials Science (PASUMS)". Since the launch of PASUMS, various open-source software programs have been developed/advanced, including ab initio calculations, effective model solvers, and software for machine learning. We also focus on activities that make the software easier to use, such as developing comprehensive computing tools that enable efficient use of supercomputers and interoperability between different software programs. We hope to contribute broadly to developing the computational materials science community through these activities.

Figures (9)

• Figure 1: Schematic of electrical conductance and eigenchannels (yellow) in a nanostructure. Periodic boundary conditions and Bloch's theorem are applied in the vertical and front-back directions. In contrast, the periodic structure of each electrode material continues semi-infinitely in the left-right direction.
• Figure 2: Illustration of the directed loop algorithm for $S=1/2$ spin model. Vertical solid lines and dashed lines denote upspin and downspin, respectively. First, vertices (dashed horizontal lines) are generated. Next, a pair of $\hat{S}^+$ (black circle) and $\hat{S}^-$ (white circle) operators are inserted. Then, $\hat{S}^+$ moves along lines while flipping spins, and returns where $\hat{S}^-$ is and removed.
• Figure 3: (a) Schematic of the ESM method. The Poisson equation can be solved analytically in regions of perfect conductors and vacuum, and these results are connected to the electrostatic potential within the simulation cell for efficient non-periodic system calculations. The Kohn-Sham equation in each self-consistent step is solved within the simulation cell under conventional periodic boundary conditions. (b) Schematic of the ESM-RISM method. The solvent density is represented in grayscale. The Kohn-Sham and RISM equations are solved in separate simulation cells, interconnected through the electrostatic potential.
• Figure 4: Schematic of downfolding. From the band structure and Kohn-Sham orbitals obtained by first-principles calculations, parameters such as hopping integrals $t$ and Coulomb integrals $U$ of the Hubbard model are calculated by focusing only on the states near the Fermi surface (target bands) using maximally localized Wannier functions and constrained random phase approximation. Contributions from orbitals other than the target bands are included as screening to the atomic potential and electron-electron Coulomb interactions.
• Figure 5: Schematic flow of calculations using H-wave. The users prepare the interaction definition files in the Wannier90 format or the expert-mode format and the input parameter files. The interaction definition files can be generated from a simple description by StdFace, or derived from the first-principles calculations. The results are stored in the output files according to the input parameters, including the expectation values of the physical observables, the Green's functions, and other data for further analyses.