English

• Confronted with the dual challenges of global energy shortage and environmental pollution, the limitations of fossil fuels force us to urgently find sustainable energy solutions [1-7]. Rechargeable metal air batteries have inestimable for optimizing energy structure and realizing energy transformation due to their advantages of high energy density and cleanliness [8-12]. However, metal-air batteries driven by Pt-catalyzed oxygen reduction reaction (ORR) and RuO₂/IrO₂-catalyzed oxygen evolution reaction (OER) are faced with dilemma problems of manufacturing cost and performance defects in practical applications [13-16]. Specifically, different materials used in anode/cathode electrodes will undoubtedly increase the process difficulty [17-20]. Therefore, development of stable, active, non-noble, bifunctional materials that simultaneously accelerate ORR/OER is of profound significance.

  Inspired by the pioneering work on cobalt phthalocyanines, the development of the graphene with transition metals and nitrogen (TM/N) moieties is vigorous due to the adjustable property and high atomic efficiency [21-26], represented by FeN₄ and CoN₄ active centers [27-35]. For example, Zeng et al. demonstrated oxygen reduction overpotentials η^ORR of 0.42 V and oxygen evolution overpotentials η^OER of 0.72 V for FeN₄ meanwhile η^ORR of 0.50 V and η^OER of 0.37 V for CoN₄, respectively [36]. Furthermore, Yang et al. discovered η^ORR of 0.64 V and η^OER of 1.02 V for FeN₄ meanwhile η^ORR of 0.76 V and η^OER of 0.93 V for CoN₄, respectively [37]. In despite of data deviation, a consensus has been achieved that oxygen electrocatalysis over FeN₄ and CoN₄ both suffer from too strong affinity toward O containing intermediates. Recalling the well-established volcano-curves, to suppress the adsorption capacity is the way to further activate the FeN₄ and CoN₄ active centers, as revealed by Scheme S1 (Supporting information).

  Our previous study has revealed the formation of spatial covalent binding between CoN₄ center and S/P/B in the interspace of graphene-based bilayer [38,39]. The presence of such interaction reduces the adsorption ability and boosts the ORR catalytic activity. Similarly, Hu et al. designed a graphene-based bilayer consisting of FeN₄ center in the upper layer and TMN₄ in the supporting layer and discovered the formation of Fe-TM bond in the confined space [40]. Their results proposed the combination of FeN₄ and HfN₄ is a trifunctional active center for ORR, OER and HER. Similar interaction in stacking graphene layers has been achieved experimentally. Zhang et al. has prepared FeN₅ configuration due to the spatial FeN₄-N interaction which obtained through prolonged calcination of melamine and hemin co-adsorbed on oxide graphene [41]. Wherein, the carbon material with FeN₅ center exhibited superior ORR activity with half-wave potential of 0.84 V compared to FeN₄ counterpart with half-wave potential of 0.72 V in alkaline media. Wu et al. has successfully synthesized the FeN₄-FeN₄ coupling through the in situ atomization of graphene supported metal oxide nanoparticles and confirmed its optimized ORR activity with a half-wave potential of 0.90 V under alkaline conditions and 0.74 V under acidic conditions [42]. The improved ORR activity stems from the weakened adsorption capacity caused by the formation of spatial covalent binding which performs as a fifth coordination. On the other hand, the influence of the inter-binding on the OER performance of FeN₄ and CoN₄ is rarely reported. Therefore, the systematical investigation focused on the influence of the inter-binding on the oxygen electrocatalysis of FeN₄ and CoN₄ is still necessary and the revealing of the key factors that control the performance is meaningful.

  In this work, we construct a graphene-based bilayer as our prototype, consisting of FeN₄ or CoN₄ embedding graphene as the upper layer and TMC₃ or TMN₃ doping graphene as the sublayer. The selection of such prototype is due to that the facts that (1) CoN₄ or FeN₄ doped graphene have been confirmed to offer superior ORR/OER activities; (2) considering the planar structure of TMN₄ doped graphene and the protrusion structure of TMN₃ or TMC₃ doped graphene [43-46], the inter-binding formation is expected. Fig. 1a schematically describes the atomic configurations. The stability and activity are evaluated by density functional theory calculations. Firstly, we perform density functional theory (DFT) calculations to identify the bifunctional activity of our designed bilayers. Subsequently, we perform machine learning analysis to figure out the key characteristics that affects the catalytic activity. This work opens up a new path for designing the efficient oxygen electrocatalysts via modification of coordination environment.

  Figure 1

  

  Figure 1.  Structural characterization. (a) Atomic structure of our designed bilayers. Herein, the atoms in grey, blue, gold, purple and orange colors stand for C, N, N/C, Co/Fe and TM elements, respectively. (b) Deformation electron distribution between dual metal atoms of CoN₄/CoC₃, CoN₄/CoN₃, FeN₄/FeC₃ and FeN₄/FeN₃, respectively. Top panel: systems without adsorption; Bottom panel: systems upon OH adsorption.

  DownLoad: Full-Size Img PowerPoint

  Fig. 1a is schematic diagram of our designed bilayer structures. Through geometry optimization, the spontaneous formation of inter-binding is observed, as supported by the small bond length L between dual metal atoms, the negative binding energy E_b between upper layer and bottom layer, and the dd orbital overlapping between dual metal atoms (Figs. S1–S3 in Supporting information). Herein, taken CoN₄/CoC₃, CoN₄/CoN₃, FeN₄/FeC₃ and FeN₄/FeN₃ as representation, the deformation electron distributions between dual metal atoms visualize such interaction, as shown in Fig. 1b. Moreover, the thermodynamic stability of such interaction has been confirmed by ab initio first molecular dynamics simulation at 300 K (Fig. S4 in Supporting information). What is more, upon the adsorption of O-involving intermediates (Figs. S5–S8 in Supporting information), the inter-binding is weakened that agrees with the common sense [38,39], as reflected by the reduced deformation electron density between dual metal atoms in Fig. 1b. However, the inter-binding is still preserved (Tables S1 and S2, Fig. S9 in Supporting information). Therefore, the axial binding is robust once it is formed in bilayers.

  The adsorption configurations of O-involving intermediates are shown in Figs. S5–S8. From the adsorption energies listed in Tables S3 and S4 (Supporting information), it is general rule that the adsorption capacity is weakened in comparison with the monolayer. Since the ORR activity of FeN₄ and CoN₄ suffers from too strong adsorption ability, it is reasonable to expect an improved activity caused by the suppressed adsorption. Herein, three points should be noted that (1) CoC₃, CoN₃, FeC₃ and FeN₃ in the supporting layer are unable to capture OH intermediate as demonstrated in Fig. S10 (Supporting information). Therefore, we do not consider the reactivity of the bottom layer in the following discussions; (2) the coordination environment poses a great influence on the adsorption capacity. For example, the OH adsorption on FeN₄/FeN₃ is weakened as compared FeC₄/FeN₃ (Table S5 and Fig. S11 in Supporting information). However, since the identified good bifunctional oxygen electrocatalysis, we only focused our attention on CoN₄ and FeN₄ sites in current work; (3) if the dual metal sites are well separated without the formation of inter-binding, the adsorption abilities of CoN₄ and FeN₄ sites also show some variation in comparison with the monolayer counterparts (Table S6 and Fig. S12 in Supporting information). However, our goal is focused on understanding the catalytic variation brought by the axial coordination, therefore, we also do not consider the separation situation currently.

  To confirm our expectation, the free energy profiles along 4e⁻ ORR/OER mechanism at the applied potential of 1.23 V are given in Fig. 2a, Figs. S13 and S14 (Supporting information). Affected by the strong toxicity, the ORR potential-determining step (PDS) of CoN₄ and FeN₄ are located at the last step of OH* + (H⁺ + e⁻) → H₂O [28,47,48]. The ORR overpotentials η^ORR of CoN₄ and FeN₄ is 0.53 and 0.76 V, respectively, which is close to the values reported in the literatures [36,49,50]. Move to bilayers, PDS is still located at the last protonation step of OH* + (H⁺ + e⁻) → H₂O for FeN₄-based bilayers shown in Table S7 (Supporting information); however, it changes to be first protonation step of O₂* + (H⁺ + e⁻) → OOH* for CoN₄-based bilayers, respectively. The η^ORR are 0.64, 0.36, 0.37 and 0.38 V for CoN₄/CoC₃, CoN₄/CoN₃, FeN₄/FeC₃ and FeN₄/FeN₃, respectively. A reduced η^ORR implies the improved ORR activity. In the regard, the activity is deteriorated for CoN₄/CoC₃ because the weak adsorption that leads to move away from the top of the volcano. However, the latter three combinations even perform better than the benchmark of Pt(111) with η^ORR of 0.43 V [36,51]. This clearly confirms our expectation that the reduced affinity into the suitable range would boost the ORR activity. Furthermore, the O* formation is energetically preferred in comparison with H₂O₂ formation, indicating the good selectivity of 4e⁻ reduction to H₂O, as shown in Table S7. According to η^ORR summary in Figs. 2b and c, we further identify a series of CoN₄-based bilayer involving CoN₄/ScC₃, CoN₄/TiC₃, CoN₄/CuC₃ and CoN₄/TiN₃ and a series of FeN₄-based bilayer involving FeN₄/MnC₃, FeN₄/CoC₃, FeN₄/ZnC₃, FeN₄/CrN₃, FeN₄/MnN₃, FeN₄/NiN₃, FeN₄/ZnN₃ are attractive toward oxygen reduction electrocatalysis due to η^ORR less than 0.43 V.

  Figure 2

  

  Figure 2.  Bifunctional activity. (a) Free energy diagram of CoN₄, CoN₄/CoC₃ and CoN₄/CoN₃, FeN₄, FeN₄/FeC₃ and FeN₄/FeN₃, respectively. (b) The ORR/OER overpotentials of FeN₄ and FeN₄/TMC₃ and FeN₄/TMN₃, respectively. (c) The ORR/OER overpotentials of CoN₄ and CoN₄/TMC₃ and CoN₄/TMN₃, respectively. (d) The theoretical OER and ORR polarization curves of CoN₄, CoN₄/CoN₃, FeN₄, FeN₄/FeC₃ and FeN₄/FeN₃, respectively.

  DownLoad: Full-Size Img PowerPoint

  For the OER side, PDS of CoN₄ and FeN₄ are separately OH* deprotonation and OOH* formation [37,47]. Move to bilayers, there is no change of PDS. Referring to an IrO₂ benchmark with the OER overpotential η^OER of −0.65 V [36,49], η^OER are −0.78, −1.03, −0.91, −0.59, −0.48, and −0.53 V for CoN₄, FeN₄, CoN₄/CoC₃, CoN₄/CoN₃, FeN₄/FeC₃ and FeN₄/FeN₃, respectively. According to η^OER summary in Figs. 2b and c, a series of CoN₄-based bilayers involving CoN₄/CrC₃, CoN₄/ScN₃, CoN₄/TiN₃, CoN₄/VN₃ and CoN₄/CrN₃ and a series of FeN₄-based bilayers involving FeN₄/ScC₃, FeN₄/TiC₃, FeN₄/MnC₃, FeN₄/CoC₃, FeN₄/NiC₃, FeN₄/CuC₃, FeN₄/ZnC₃, FeN₄/ScN₃, FeN₄/TiN₃, FeN₄/CrN₃, FeN₄/MnN₃, FeN₄/CuN₃ and FeN₄/ZnN₃ are also expected to be effective.

  Herein, the parameter of Δη is considered to characterize the capacity of bifunctional oxygen electrocatalysis wherein Δη is defined as the difference between η^OER and η^ORR (Fig. S15 in Supporting information) [52,53]. The Δη of CoN₄, FeN₄, CoN₄/CoC₃, CoN₄/CoN₃, FeN₄/FeC₃ and FeN₄/FeN₃ were calculated to be 1.31, 1.79, 1.55, 0.95, 0.85 and 0.91 V, respectively. Among them, the ∆η of CoN₄/CoN₃, FeN₄/FeC₃ and FeN₄/FeN₃ are smaller than that of 1.08 V for the combination of Pt and IrO₂, indicating the potential to be high-performance reversible oxygen electrodes. Moreover, to visualize the catalytic performances of the screened out CoN₄/CoN₃, FeN₄/FeC₃ and FeN₄/FeN₃, their theoretical OER and ORR polarization curves are simulated in Fig. 2d. The voltage difference ∆E between half-wave potential E_1/2 of ORR and the onset potential E_onset of OER at a current density of 10 mA/cm² reduces from 0.82 V of CoN₄ to 0.60, 0.42 and 0.47 V of CoN₄/CoN₃, FeN₄/FeC₃ and FeN₄/FeN₃, respectively. Therefore, it clearly supports the good activity toward bifunctional oxygen electrocatalysis when suitable bilayer is constructed in comparison with single layer counterpart. What is more, a series of FeN₄-based bilayers involving FeN₄/CoC₃, FeN₄/NiC₃, FeN₄/ZnC₃, FeN₄/CrN₃ and FeN₄/ZnN₃, have been identified as bifunctional oxygen electrocatalysts (Tables S8 and S9, Fig. S16 in Supporting information). Before closing activity discussions, we would like to mention that the similar spatial five-coordination structures confined in bilayers have been synthesized, such as, the FeN₄-N interaction [41], the FeN₄-FeN₄ coupling [42,54,55], and FeN₄-NiN₄ combination [54]. Their results give us the confidence that our designed bilayers would be achieved considering the recent advance in synthesis technologies.

  To clarify the relationship between adsorption and catalytic activity, we perform varied linear fittings and confirm that the oxygen electrocatalysis stems from the adsorption capacity (Fig. S17 in Supporting information). Interestingly, for either ORR or OER volcano plot, FeN₄-based bilayers are mostly located at the strong branch and CoN₄-based bilayers are on the weak branch. More importantly, the optimal E_ads(OH*) for bifunctional oxygen electrocatalysis should be in the range from −2.8 eV to −2.3 eV. However, based on the data listed in Tables S10–S12 (Supporting information), there are no promising candidates for bifunctional oxygen electrode where the rest 3d TM with exception of Co and Fe are taken into consideration.

  To explore the physical origin of boosted activity, partial density of states (PDOS) is given in Fig. 3, Figs. S18 and S19 (Supporting information). Compared with the monolayer counterpart, the d electron distributions of Co or Fe in the upper layer are significantly tuned by the inter-binding in bilayer. Therefore, the reactivity change between bilayer and monolayer is reasonable. What is more, Fe has the higher d band that interacted with adsorbate OH* referring to Fermi energy level in comparted with Co d band, as indicative of d band center ε (Tables S3 and S4). That is in line with the volcano plots featured with FeN₄-based bilayers at the strong branch and CoN₄-based bilayers at the weak branch. In the regard, we consider the linear fitting between ε and E_ads(OH*) and ΔG_OH* (Fig. S20 in Supporting information). Roughly, the upshifted ε leads to the enhanced adsorption. However, we also notice a large deviation.

  Figure 3

  

  Figure 3.  Electronic Analysis. (a) PDOS between d band of Co in upper layer and d band of Co in bottom layer of CoN₄, CoN₄/CoC₃ and CoN₄/CoN₄. (b) PDOS between d band of Fe in upper layer and d band of Fe in bottom layer of FeN₄, FeN₄/FeC₃ and FeN₄/FeN₃. (c) Mulliken charges of TM in upper layer. (d) PDOS between d band of Co in upper layer and p band of adsorbate OH* of CoN₄, CoN₄/CoC₃ and CoN₄/CoN₄ upon OH adsorption. (e) PDOS between d band of Co in upper layer and p band of adsorbate OH* of FeN₄, FeN₄/FeC₃ and FeN₄/FeN₃. (f) Mulliken charges of TM in upper layer and Mulliken charges of adsorbate OH* upon OH adsorption. Wherein the orange and green stand for the d band of TM in upper layer and bottom layer, respectively.

  DownLoad: Full-Size Img PowerPoint

  Since the classic d band theory does not take the ionic binding into consideration which may raise up the scatter as observed [38,39], therefore, we further analyze the charge distributions in order to character electrostatic interaction in addition to covalent interaction. The Mulliken charges are given in Fig. 3 and Figs. S21–S23 (Supporting information). Clearly, the inter-binding formation significantly enriches the electron accumulation on Co or Fe in the upper layer, as indicated by negative Mulliken charge of bilayers vs. positive Mulliken charge of monolayer. Upon OH adsorption, although Mulliken charge of adsorbate OH* is almost unchangeable, Mulliken charge of Co or Fe in the upper layer is significantly tuned by the TM selection in the bottom layer. For instance, there is electrostatic repulsion between the CoN₄/CoC₃ and adsorbate OH*; conversely, electrostatic attractions are found for CoN₄/CoN₃, FeN₄/FeC₃ and FeN₄/FeN₃ with OH adsorption. Therefore, the varied electrostatic interaction is one important reason for the relative discretion observed.

  To enclose our discussions, we perform machine learning (ML) to uncover the decisive factor that guides the adsorption mode of our designed bilayers, deeper analysis focused on E_ads(OH*) is accomplished through three diverse ML algorithms: LASSO, RFR, and XGBoost due to their universal application in the field of electrocatalysis [56-58]. The E_ads(OH*) of varied bilayers are listed in Tables S10 and S11 (Supporting information) as output. At the beginning 13 features are used as input (Table S13 in Supporting information). Pearson correlation coefficient (PCC) between pairwise feature is evaluated to eliminate the highly redundant features in Fig. 4a. Herein, we artificially select the value of 0.5 to accelerate the compromise between the prediction speed and the prediction accuracy of the machine learning progress. Considering |PCC| ≤ 0.5, the features 0, 1, 2, 4, 6, 8, 10, 11 and 12 are employed as the input features for subsequent ML models. According to R² and RMSE in Fig. 4b, LASSO model has poor generalization ability whereas the RFR and XGBoost models deliver relatively good performance featured with the R²/RMSE of 0.75/0.48 and 0.83/0.35, respectively. According to the fitted scatter diagram between the adsorption energies E_ads^DFT calculated by DFT and E_ads^ML predicted by XGBoost model in Fig. 4c, the good linear character further indicates its applicability in our designed bilayers. Thereby, we analyze the importance of the input features in Fig. 4d. Evidently, the binding energy E_b (feature 0) is identified as the most important feature accounting for 46.5%. Furthermore, the instinct chemical-physical property of atom acted as active site is also vital, as reflected by 22.9% of R_TM-up (feature 2), 4.9% V_TM-up (feature 4), 8.6% of N_TM-up (feature 6), 6.3% Χ_TM-up (feature 8). Combining DFT analysis with ML analysis, the bifunctional activity of TMN₄-based bilayer stems from the inter-binding strength caused by axial coordination and the instinct chemical-physical property of the specifical atom acted as the active site. As consequence, our work provides a vital but simple strategy to promote the catalytic activity via constructing axial inter-binding.

  Figure 4

  

  Figure 4.  Machining Learning. (a) Pearson correlation coefficient between features. (b) The RMSE and R² with LASSO, RFR and XGBoost algorithms. (c) Comparison of DFT calculated E_ads(OH*) with predicted by XGBoost method. (d) The feature importance analysis of XGBoost method.

  DownLoad: Full-Size Img PowerPoint

  In conclusion, we propose an axial modulation of coordination environment to realize highly effective oxygen electrocatalysis over FeN₄ or CoN₄-based bilayer. Our results show that the inter-binding formed between the graphene-based layers enhances the structural stability, weakens the adsorption affinity and improves the bifunctional activity. To be specific, a series of FeN₄-based bilayers, involving FeN₄/FeC₃, FeN₄/CoC₃, FeN₄/NiC₃, FeN₄/ZnC₃, FeN₄/CrN₃, FeN₄/FeN₃, FeN₄/ZnN₃ are attractively promising in application of metal-air battery. According to DFT/ML analysis, bifunctional activity of TMN₄-based bilayer origins from the construction of axial coordination and the instinct reactivity of the active site. Overall, our work provides theoretical guidance for construction highly active single atom catalysts via adding axial inter-binding.

  Declaration of competing interest

  The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

  CRediT authorship contribution statement

  Ze Zhang: Investigation. Lei Yang: Methodology. Jin-Ru Liu: Investigation. Hao Hu: Methodology. Jian-Li Mi: Methodology. Chao Su: Formal analysis. Bei-Bei Xiao: Writing – review & editing, Writing – original draft, Formal analysis. Zhi-Min Ao: Writing – review & editing, Supervision.

  Acknowledgments

  The authors greatly acknowledge the financial support from the National Natural Science Foundation of China (Nos. 21503097, 52130101, 51701152, 21806023, and 51702345), Postgraduate Research & Practice Innovation Program of Jiangsu Province (No. KYCX23_3905).

  Supplementary materials

  Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.cclet.2024.110013.

  
  
• 1. [1]

     Q.E. Huang, B.L. Wang, S. Ye, et al., ACS Catal. 12 (2022) 491–496. doi: 10.1021/acscatal.1c04644

  2. [2]

     T.Y. Liu, Y. Wang, Y.F. Li, Adv. Funct. Mater. 32 (2022) 2207110. doi: 10.1002/adfm.202207110

  3. [3]

     C.H. Qi, H.Y. Yang, Z.Q. Sun, et al., Angew. Chem. Int. Ed. 62 (2023) e202308344. doi: 10.1002/anie.202308344

  4. [4]

     A. Yu, W.T. Long, L.T. Zhu, et al., Chin. Chem. Lett. 34 (2023) 107860. doi: 10.1016/j.cclet.2022.107860

  5. [5]

     Z. Li, X.Q. Qi, J.J. Wang, et al., SusMat 3 (2023) 498–509. doi: 10.1002/sus2.148

  6. [6]

     W.J. Zhang, S.S. Liu, Y. Yang, et al., Angew. Chem. Int. Ed. 62 (2023) e202219241. doi: 10.1002/anie.202219241

  7. [7]

     X. Huang, K.Y. Wu, C. Su, et al., Chin. Chem. Lett. (2024) 109720. https://doi.org/10.1016/j.cclet.2024.109720. doi: 10.1016/j.cclet.2024.109720

  8. [8]

     J.J. Ban, X.H. Wen, H.J. Xu, et al., Adv. Funct. Mater. 31 (2021) 2010472. doi: 10.1002/adfm.202010472

  9. [9]

     H. Yang, S. Gao, D.W. Rao, X.H. Yan, Energy Storage Mater. 46 (2022) 553–562. doi: 10.1016/j.ensm.2022.01.040

  10. [10]

      J. Zhang, Y.F. Zhao, W.T. Zhao, et al., Angew. Chem. Int. Ed. 62 (2023) e202314303. doi: 10.1002/anie.202314303

  11. [11]

      S. Xie, H.C. Jin, C. Wang, et al., Chin. Chem. Lett. 34 (2023) 107681. doi: 10.1016/j.cclet.2022.07.024

  12. [12]

      J. Zhang, Y. Chen, M. Tian, et al., Chin. Chem. Lett. 34 (2023) 107886. doi: 10.1016/j.cclet.2022.107886

  13. [13]

      F.X. Ma, Z.Q. Liu, G.B. Zhang, et al., Small 19 (2023) 2207991. doi: 10.1002/smll.202207991

  14. [14]

      X. Wang, Y. An, L.F. Liu, et al., Angew. Chem. Int. Ed. 61 (2022) e202209746. doi: 10.1002/anie.202209746

  15. [15]

      Y.M. Tian, H. Wu, A. Hanif, et al., Chin. Chem. Lett. 34 (2023) 108056. doi: 10.1016/j.cclet.2022.108056

  16. [16]

      H. Wang, L.Y. Cao, Y.Q. Feng, et al., Chin. Chem. Lett. 34 (2023) 107601. doi: 10.1016/j.cclet.2022.06.024

  17. [17]

      P. Zhang, K. Chen, J.Y. Li, et al., Adv. Mater. 35 (2023) 2303243. doi: 10.1002/adma.202303243

  18. [18]

      Q. Sun, L. Dai, T.T. Luo, et al., Carbon Energy 5 (2023) e276. doi: 10.1002/cey2.276

  19. [19]

      Q. Lu, X.H. Zou, Y.F. Bu, Z.P. Shao, Energy Storage Mater. 55 (2023) 166–192. doi: 10.1016/j.ensm.2022.11.046

  20. [20]

      T.S. Bai, D.P. Li, S.Y. Xiao, et al., Energy Environ. Sci. 16 (2023) 1431–1465. doi: 10.1039/d2ee02949a

  21. [21]

      J.C. Liu, F. Luo, J. Li, J. Am. Chem. Soc. 145 (2023) 25264–25273. doi: 10.1021/jacs.3c08697

  22. [22]

      C.Y. Zhang, Y.H. Dai, Q. Sun, et al., Mater. Today Adv. 16 (2022) 100280. doi: 10.1016/j.mtadv.2022.100280

  23. [23]

      X.M. Zhang, Z.X. Xia, H.Q. Li, et al., J. Colloid Interface Sci. 640 (2023) 170–178. doi: 10.1117/12.2689877

  24. [24]

      H.M. Liu, Q.L. Liu, Y.R. Wang, et al., Chin. Chem. Lett. 33 (2022) 683–692. doi: 10.1016/j.cclet.2021.07.038

  25. [25]

      Y.A. Shang, X.G. Duan, S.B. Wang, et al., Chin. Chem. Lett. 33 (2022) 663–673. doi: 10.1016/j.cclet.2021.07.050

  26. [26]

      W.H. Zhong, Y. Qiu, H.J. Shen, et al., J. Am. Chem. Soc. 143 (2021) 4405–4413. doi: 10.1021/jacs.1c00889

  27. [27]

      J.J. Liu, Z.X. Wei, Z.C. Gong, et al., Appl. Catal. B: Environ. 324 (2023) 122267. doi: 10.1016/j.apcatb.2022.122267

  28. [28]

      S.Y. Chen, T. Luo, X.Q. Li, et al., J. Am. Chem. Soc. 144 (2022) 14505–14516. doi: 10.1021/jacs.2c01194

  29. [29]

      Y.Q. Wang, K. Liu, J. Li, et al., Chem. Eng. J. 429 (2022) 132119. doi: 10.1016/j.cej.2021.132119

  30. [30]

      L. Huang, Q. Liu, W.W. Wu, et al., Nat. Commun. 14 (2023) 5594. doi: 10.1038/s41467-023-41311-9

  31. [31]

      F.X. Zhang, X.P. Liu, Y. Chen, et al., Chin. Chem. Lett. 34 (2023) 108142. doi: 10.1016/j.cclet.2023.108142

  32. [32]

      R.Q. Zhong, C.X. Zhi, Y.X. Wu, et al., Chin. Chem. Lett. 31 (2020) 1588–1592. doi: 10.1016/j.cclet.2019.12.004

  33. [33]

      B.B. Xiao, L. Yang, H.Y. Liu, et al., Appl. Surf. Sci. 537 (2021) 147846. doi: 10.1016/j.apsusc.2020.147846

  34. [34]

      X. Hu, S.Y. Chen, L.T. Chen, et al., J. Am. Chem. Soc. 144 (2022) 18144–18152. doi: 10.1021/jacs.2c08743

  35. [35]

      K.M. Zhao, S.Q. Liu, Y.Y. Li, et al., Adv. Energy Mater. 12 (2022) 2103588. doi: 10.1002/aenm.202103588

  36. [36]

      H.X. Xu, D.J. Cheng, D.P. Cao, X.C. Zeng, Nat. Catal. 1 (2018) 339–348. doi: 10.1038/s41929-018-0063-z

  37. [37]

      A.Z. Yang, K.Y. Su, W. Lei, et al., Adv. Energy Mater. 13 (2023) 2203150. doi: 10.1002/aenm.202203150

  38. [38]

      B.B. Xiao, Z. Zhang, L.B. Yu, et al., Surf. Interfaces 33 (2022) 102232. doi: 10.1016/j.surfin.2022.102232

  39. [39]

      L.B. Yu, Q.Y. Huang, J. Wu, et al., Chin. J. Chem. Eng. 54 (2023) 106–113. doi: 10.1016/j.cjche.2022.03.007

  40. [40]

      R.M. Hu, Y.C. Li, F.H. Wang, J.X. Shang, Nanoscale 12 (2020) 20413–20424. doi: 10.1039/d0nr05202g

  41. [41]

      H.N. Zhang, S.P. Jia, X.L. Shi, et al., Mater. Chem. Front. 5 (2021) 8127–8137. doi: 10.1039/d1qm00809a

  42. [42]

      X. Wu, Q.C. Wang, S.Z. Yang, et al., Energy Environ. Sci. 15 (2022) 1183–1191. doi: 10.1039/d1ee03311e

  43. [43]

      Q.K. Li, X.F. Li, G.Z. Zhang, J. Jiang, J. Am. Chem. Soc. 140 (2018) 15149–15152. doi: 10.1021/jacs.8b07816

  44. [44]

      S.X. Luo, M. Lei, H.X. Ren, et al., Surf. Sci. 720 (2022) 122049. doi: 10.1016/j.susc.2022.122049

  45. [45]

      Y. Xiong, H.C. Li, C.W. Liu, et al., Adv. Mater. 34 (2022) 2110653. doi: 10.1002/adma.202110653

  46. [46]

      R.X. Zhang, N. Shehzad, L.X. Zhang, et al., Chem. Phys. 570 (2023) 111888. doi: 10.1016/j.chemphys.2023.111888

  47. [47]

      H.L. Fei, J.C. Dong, Y.X. Feng, et al., Nat. Catal. 1 (2018) 63–72. doi: 10.1038/s41929-017-0008-y

  48. [48]

      X.L. Hu, N.Q. Su, J. Phys. Chem. Lett. 14 (2023) 9872–9882. doi: 10.1021/acs.jpclett.3c02412

  49. [49]

      R.H. Lu, C.X. Quan, C.Y. Zhang, et al., Nano Res. 15 (2022) 6067–6075. doi: 10.1007/s12274-022-4318-2

  50. [50]

      Z.Q. Tian, Y. Wang, Y.L. Li, et al., iScience 25 (2022) 104557. doi: 10.1016/j.isci.2022.104557

  51. [51]

      B.B. Xiao, H.Y. Liu, L. Yang, et al., ACS Appl. Energy Mater. 3 (2020) 260–267. doi: 10.1021/acsaem.9b01457

  52. [52]

      N.G. Ma, Y.Q. Zhang, Y.H. Wang, et al., Appl. Surf. Sci. 628 (2023) 157225. doi: 10.1016/j.apsusc.2023.157225

  53. [53]

      T.Y. Yan, X.Y. Li, Z.X. Wang, et al., J. Colloid Interface Sci. 649 (2023) 1–9. doi: 10.1016/j.jcis.2023.06.066

  54. [54]

      L.X. Wang, X.P. Gao, S.C. Wang, et al., J. Am. Chem. Soc. 145 (2023) 13462–13468. doi: 10.1021/jacs.3c04172

  55. [55]

      X.D. Shi, Z.H. Pu, B. Chi, et al., Small 20 (2023) 2307011.

  56. [56]

      B.W. Deng, P. Chen, P. Xie, et al., Chem. Eng. Sci. 267 (2023) 118368. doi: 10.1016/j.ces.2022.118368

  57. [57]

      S. Kapse, N. Barman, R. Thapa, Carbon 201 (2023) 703–711. doi: 10.1016/j.carbon.2022.09.059

  58. [58]

      L. Yang, J.K. Fan, B.B. Xiao, W.H. Zhu, Chem. Eng. J. 468 (2023) 143823. doi: 10.1016/j.cej.2023.143823

• 

  Figure 1  Structural characterization. (a) Atomic structure of our designed bilayers. Herein, the atoms in grey, blue, gold, purple and orange colors stand for C, N, N/C, Co/Fe and TM elements, respectively. (b) Deformation electron distribution between dual metal atoms of CoN₄/CoC₃, CoN₄/CoN₃, FeN₄/FeC₃ and FeN₄/FeN₃, respectively. Top panel: systems without adsorption; Bottom panel: systems upon OH adsorption.

  下载: 全尺寸图片 幻灯片
  

  Figure 2  Bifunctional activity. (a) Free energy diagram of CoN₄, CoN₄/CoC₃ and CoN₄/CoN₃, FeN₄, FeN₄/FeC₃ and FeN₄/FeN₃, respectively. (b) The ORR/OER overpotentials of FeN₄ and FeN₄/TMC₃ and FeN₄/TMN₃, respectively. (c) The ORR/OER overpotentials of CoN₄ and CoN₄/TMC₃ and CoN₄/TMN₃, respectively. (d) The theoretical OER and ORR polarization curves of CoN₄, CoN₄/CoN₃, FeN₄, FeN₄/FeC₃ and FeN₄/FeN₃, respectively.

  下载: 全尺寸图片 幻灯片
  

  Figure 3  Electronic Analysis. (a) PDOS between d band of Co in upper layer and d band of Co in bottom layer of CoN₄, CoN₄/CoC₃ and CoN₄/CoN₄. (b) PDOS between d band of Fe in upper layer and d band of Fe in bottom layer of FeN₄, FeN₄/FeC₃ and FeN₄/FeN₃. (c) Mulliken charges of TM in upper layer. (d) PDOS between d band of Co in upper layer and p band of adsorbate OH* of CoN₄, CoN₄/CoC₃ and CoN₄/CoN₄ upon OH adsorption. (e) PDOS between d band of Co in upper layer and p band of adsorbate OH* of FeN₄, FeN₄/FeC₃ and FeN₄/FeN₃. (f) Mulliken charges of TM in upper layer and Mulliken charges of adsorbate OH* upon OH adsorption. Wherein the orange and green stand for the d band of TM in upper layer and bottom layer, respectively.

  下载: 全尺寸图片 幻灯片
  

  Figure 4  Machining Learning. (a) Pearson correlation coefficient between features. (b) The RMSE and R² with LASSO, RFR and XGBoost algorithms. (c) Comparison of DFT calculated E_ads(OH*) with predicted by XGBoost method. (d) The feature importance analysis of XGBoost method.

  下载: 全尺寸图片 幻灯片
•