第146回 第3部

Quantum chemistry software comprises immensely useful tools in material and biological science research. Widely diverse programs have been developed in Western countries as Asian countries including Japan have lagged. In fact, only a few programs have been developed in Japan. The mission of our research team is to provide K computer users with a high-performance software for quantum molecular simulation. In the early stage of the K computer project, no quantum chemistry software was available for general purpose and massively parallel computation on the K computer because not every program was designed for use on it. Therefore, we have chosen to develop a new comprehensive ab initio quantum chemistry software: NTChem [1, 2]. It is completely new software that implements not only standard quantum chemistry approaches, but also original and improved theoretical methods that we have developed in our research work. Currently, NTChem is available on several supercomputer and PC cluster systems as well as the K computer. In the near future, we hope to make NTChem available to the general public. We intend to continue adopting users’ requests with the aim of making the program more convenient and user-friendly for researchers in various fields. We earnestly hope that NTChem will be an important tool leading the way toward a new frontier of computational molecular science.
In this presentation, we will present an overview of NTChem software and its applications. In particular, we will give a talk about the next version of NTChem. In standard quantum chemical codes, the arrays of the Fock and density matrices are not separated per CPU core, that is, all CPU cores have full-size matrices. Since these array sizes increase with the square of the molecular size, we often suffer from the memory overflow in large molecular systems. To overcome this problem, we are developing the memory-distributed NTChem with the massively-parallel sparse-matrix library NTPoly. In this talk, we present that our developed code enables to calculate large molecular system consisting of more than 10,000 atoms on the K computer. We also show that the parallel efficiencies, which are 59% for 8,192 nodes and 39% for 16,384 nodes with reference to 2,048 nodes are obtained using MPI/OpenMP hybrid parallel computations based on the DFT.
(1) NTChem.
(2) T. Nakajima, M. Katouda, M. Kamiya, Y. Nakatsuka, Int. J. Quantum Chem. 115, 349–359 (2015).