By introducing a new basis-function contraction scheme, the multisite method, we have successfully reduced the computational cost of first-principles calculations without losing their accuracy. This method enables large-scale first-principles calculations of systems containing several thousand atoms or more for a wide range of materials, including metals, semiconductors, and insulators (upper left figure).

Using this approach, we are investigating, for example, the structures, electronic properties, and reactivity of metal nanoparticle catalysts with sizes of several nanometers, comparable to those used in practical applications (upper right figure). Since large-scale calculations generate highly complex data, we use statistical analysis and machine learning to efficiently identify characteristic sites and features within large systems. For example, statistical analysis of large-scale DFT results has enabled us to objectively and quantitatively elucidate the effects of supports and support defects on the electronic structures of nanoparticles (lower figure).

Through this approach, we aim to achieve accurate and efficient theoretical analysis of materials at the nanoscale.