(202a) Machine Learning Approach to First-Principles Database for Designing Active Nanomaterials for Electrochemical Energy Convergence

Machine learning approach has been intensively applied to nanomaterials of high activity for energy convergence, due to its evident power to identify a key correlation in database. The success of the method, therefore, is critically dependent on the reliability and accuracy of accumulated data. Here, the acquisition of crystal-clear structure-performance relation in heterogeneous catalysis of sluggish electrochemical reactions is demonstrated through three-dimensional tracking of nanoparticles in liquid medium and over thermal treatment. For typical catalyst design process using colloidal synthesis we acquire atomic positions of ligand-protected platinum nanoparticles using first-principles density functional theory calculations and in-situ TEM observation. The structural transformation in time domain is shown by decoupling and tracking of each atomic coordinate using molecular dynamics simulations standing on machine-learning potentials. Fast and accurate acquisition of structural data is integrated with a computational catalysis framework to identify a clear structure-activity correlation in the CO oxidation reaction as an example.