A Machine Learning Aided Study Of Oxygen Electrocatalysis Reactions On 2D Cabon-based Materials And QSPR Modeling

Due to their adjustable activity,it is confirmed by both experiments and DFT calculations that graphene-based metal-free and transition metal single atom catalysts have shown excellent catalytic performance in ORR and OER processes,and are applying as the cathodes in matal-air baterries.Graphene-based materials are stable in both acid and alkine medias with long cycle-life.Their adjustable activity is consequent to their adjustable charge density and conductivity.Edge modification,heteroatoms,defects(including Stone-Wales defects and single vacancy)could all contribute to such activity adjustment.Additonally,it is also found that M1M2-N-C structures with two metal active sites show great activity in ORR and OER processes as well.The synergic effects originated from the interactions between the two metal sites and between matal and the structure make changes on the surface charge density and electronic structures.Aiming for materials screening,activity descriptors and rational material design principles,oxygen electrocatalytic investigations through DFT calculations on metal-free graphene and non-noble metal M1M2-N-C structures are conducted separately.QSPR modeling is applied sequently from the structure features using graphs and chemical properties.1.Indroducing 5,7-Stone-Wales defect,single vacancy and edge modifications to nitrogen doped,nitrogen-phosphorus co-doped and nitrogen-sulfur co-doped graphene structures,a systematic investigation on ORR activity is conducted.pz-band center,an intrinsic descriptor based on the pz-obital density of the carbon active sites is raised sequently.ORR overpotential,together with the adsorption energy of the intermediate OH*and the pz-band center of the carbon sites,a QSPR model from structure features and bonds including all atoms within 4.5 A is built.It is found that ORR overpotential as the prediction target of QSPR model with a random forest algorithm,whose n_estimators=300 has the lowest mean average error and an ideal mean square error,showing the most predictivity.Based on this model,the carbon sites with lowest ORR overpotential and their neighboring atoms are explored in periodic graphene nanosheets and non-periodic clusters,their ORR overpotential being 0.398 V and 0.373 V,respectively.2.Based on graphene with 4 vacancies,3d transition metals(Ti,Mn,Fe,Co,Ni,Cu,Zn)and Ca are introduced to create M1M2-N-C structures.As the result of DFT calculations,N8V4 configurations show much more activity in ORR and OER than N6V4.Boron,carbon,phosphorus and sulfur atoms are then introduced to N8V4 configurations substituting nitrogen,and N7X and N6X2 configurations are thus created.For ORR and OER processes,two descriptors are found for predicting the catalytic performance of N8V4 configurations.The QSPR model with a random forest algorithm with 100 n_estimators has confirmed that,for ORR catalytic perforamance,the adsorption energy of OH*show same predictivity as ORR overpotential.To predict the ORR activity with the adsorption energy of OH*can save DFT calculation costs for up to 60%.Three structures with excellent bifunctional ORR/OER catalytic performance in N8V4 configuration are found,among which MnNiN7B(ηORR=0.054 V,ηOER=0.078 V)is the most active one.Formation energy is also calculated for all those N8V4 configurations to acquire a predictive QSPR model.From the prediction results of this model,all the formation energy of N8V4 are predicted minus,which means all the N8V4 configurations are possible to be synthesized and thermodynamicly stable.