Simple Descriptor Derived from Symbolic Regression Accelerating the Discovery of New Perovskite Catalysts

Wan-Jian Yin*

School of Energy, Soochow Institute for Energy and Materials InnovationS (SIEMIS), Soochow University

ABSTRACT: Descriptor is a simple scale to describe complex phenomena. In the field of catalysis, scientists have been searching for simple and accurate descriptors for decades, trying to quantitatively describe complex catalytic phenomena. These descriptors are based on the physical and chemical properties of the material, and the representative ones are free energy, d-band center, etc. Although they provide important guidance for understanding and designing catalytic materials, they all require precise experimental measurements or density functional theory (DFT) calculations, which are time-consuming and labor-intensive, and are not conducive to the high-throughput design and screening of materials.

In this report, we take the oxygen evolution reaction of oxide perovskite as an example, propose the use of symbol regression machine learning method [1], skip the DFT calculation, directly establish the catalytic activity and simple material parameters (chemical ratio, ionic radius, Electronegativity, valence, number of transition metal ions d electrons, etc.) structure-activity relationship, found a new descriptor μ/t for oxygen evolution reaction. Based on this, the material design was carried out, 13 kinds of materials were selected from more than 3000 kinds of materials, and 5 kinds of materials were successfully synthesized in the experiment. Among them, the catalytic activities of 4 new materials (Cs_0.4La_0.6Mn_0.25Co_0.75O₃, Cs_0.3La_0.7NiO₃, SrNi_0.75Co_0.25O₃ and Sr_0.25Ba_0.75NiO₃) are higher than the typical oxide perovskite catalyst BSCF, of which Cs-based oxide perovskite was reported for the first time as a catalyst for oxygen evolution reaction.

Keywords: Symbolic Regression; Machine Learning; Perovskites; Catalysis