Multi-scale simulation method to explore the relationship between structure and activity of Pt-Ni alloy catalyst
Yue Li, Jia Li*
Institute of Materials Research, Tsinghua Shenzhen International Graduate School, Shenzhen, 518000, China

EXTENDED ABSTRACT: In proton exchange membrane fuel cell (PEMFC), the overall performance of thebattery is limited by the relatively slow oxygen reduction reaction (ORR). Therefore, ORR catalyst materialplays an important role in improving the performance of PEMFC, and the Pt-Ni alloy catalyst system is one ofthe promising material systems. Pt-Ni alloy catalyst system has a variety of nanostructures, and structuralevolution occurs in the working condition. It’s difficult to obtain the atomic structure of these nanocatalysts byexperimental method. So there’s no specific explanation of the relationship between the structure and theoverall catalytic activity. This research considers to study the relationship between the microstructure and theoverall performance of Pt-Ni alloy catalyst by the multi-scale simulation method combining kinetic MonteCarlo Method (KMC) and density functional theory (DFT). We use graphs represent alloy nanoparticlestructures and generate structure space with large structural gap by graph similarity measured by graph kernel.Then active learning method is used to filter out effective train-set structures for Pt-Ni machine learningpotential (MLP). The energy prediction accuracy of the model reached 0.01-0.02 eV/atom compared to theDFT calculation results, which is improved by 30% compared with model trained by the randomly generatedstructural space. The fitted MLP is used to provide atomic interaction in the KMC process, where we studiedthe structural evolution caused by surface Ni atoms dissolution of the octahedral nanoparticles with differentcontent (Pt3Ni2, Pt1Ni1, Pt2Ni3) respectively. The evolution of nanoparticle structure, the final structure andcomponent of particles match well with the experimental results, and there is a significant difference betweennanoparticles with different content. At the same time, an OH adsorption energy dataset of the Pt catalytic sitewith different coordination environment are obtained by DFT calculations to fit a descriptor consistsofstructure features. Finally, we hope that the catalyst microstructure obtained by the KMC method combinedwith the structure descriptor of adsorption energy fitted based on DFT calculation dataset can give a goodevaluation of the overall catalytic activity of Pt-Ni catalyst nanostructures by statistical methods.