A Machine Learning Study Of The Adsorption Energy Of O₂ On Metal And Nitrogen Doped Carbon Single-Atom Catalysts

The development of clean and renewable energy sources is one major challenge faced by today’s world.Proton-exchange membrane fuel cells(PEMFCs)are considered one of the most promising solutions to the energy problem,but the sluggish kinetics of the oxygen reduction reaction(ORR),the cathode reaction in PEMFCs,severely limited their applications.Therefore,it is of vital importance to develop catalysts with high ORR activities.However,traditional Pt-based catalysts suffer a disadvantage in the high price of platinum,and this problem prompted researchers to search for transition metal catalysts that can replace precious metal based catalysts.In recently years,metal and nitrogen doped carbon single-atom catalysts(M-N-C SACs)has drawn increasing attention as a novel class of ORR catalysts with the potential to substitute Pt-based catalysts,due to their low costs,high ORR activities,and high rates of metal atom utilization.One key descriptor of a catalyst’s catalytic activity is the adsorption energy of the adsorbed molecule on it.An adsorption too strong or too weak usually indicates a weaker catalytic activity,and the best catalytic activity is usually achieved when the adsorption energy of the adsorbed molecule falls within an optimal range.For this reason,it is an important task to establish a theoretical model capable of describing the adsorption energies of reactant molecules.In this work,we conducted first-principles calculations of the adsorption energies of O2 on a variety of M-N-C SACs,and the results are analyzed using machine-learning methods.The contents of the individual chapters in this article are as follows:In the first chapter,we will give an introduction about the background information relevant to our work.We will introduce first the reaction mechanism and catalysts of ORR,then the important synthesis methods,characterization methods,and applications of SACs.In the second chapter,we will introduce the theoretical methods used in our work.The chapter will include brief introductions on Hartree-Fock equations,the basic principles of density functional theory(DFT),important exchange-correlation functionals,commonly used quantum chemistry packages,and machine-learning methods.In the third chapter,we will introduce our work on the first-principles calculations and machine-learning analysis of the adsorption energies of O2 on various M-N-C SACs.Based on the adsorption energy data of O2 on Ni(Ⅱ),Co(Ⅱ),Cu(Ⅱ),Fe(Ⅱ),Fe(III),Mn(II)single-atom catalysts supported on 15 different N-C substrates under various spin states,we discovered a good linear relationship Eads=kx+b between the adsorption energy of O2,Eads,and the data-driven descriptor x=el03/(l1-l0)(l2-l0).The descriptor only contains information about the geometrical configuration of the adsorbate(including the metal atom and the adsorbed O2 molecule),and the parameters in the linear expression,k and b,are found to contain only substrate-specific information.Our study sheds light on the respective roles played by the substrate and the adsorbate in determining the adsorption energy,and opens up possibilities for new approaches to the rational design of single-atom catalysts.