Presentation Information
[23p-21B-13]Construction of Machine Learning Potential for Titanium Oxides with Crystallographic
Shear Structure
〇(D)XUAN DAI1, Araidai Masaaki1,2, Takuma Shiga3, Toru Ujihara1,2, Shunta Harada1,2 (1.Nagoya Univ., 2.IMaSS, Nagoya Univ., 3.AIST)
Keywords:
molecular dynamics,thermal conductivity,machine-learned potential
Artificial superlattice structures have recently shown great interest in manipulating the macroscopic thermal properties of materials through nanostructure design. However, the imperfections of interfaces within artificial superlattices limit further manipulation of thermal conductivity. Therefore, we focus on titanium oxide natural superlattices with crystallographic shear (CS) planes (TinO2n-1) exhibiting extremely low interface roughens [1]. With CS planes parallel to (121)rutile or (132)rutile, the atomic arrangement deviates from rutile TiO2 by a specific shear vector 1/2[0–11]rutile. However, the intricate structural complexity of this configuration results in high computational and experimental costs, posing challenges to the in-depth investigation of the relationship between thermal conductivity and the arrangement of CS planes. The present study constructs a machine learning potential (MLP) for titanium oxides with CS structure through the combined training dataset comprising rutile TiO2 and Ti4O7 with the simplest crystal structure among the CS structures. This approach enables efficient and precise prediction of phonon transport properties.