Gibbs free energy change for the first protonation step versus additional charge on adsorbed N2 for (a) BAs, (b) BP, and (c) BSb. The blue and red curves show free energy changes for NRR at 0 and applied potentials.
- Background
- NRR Mechanisms
- Theoretical Calculations for N 2 Reduction
- References
There are two main mechanisms for NRR: a dissociative and an associative mechanism, as shown in Figure 1.2.1. The associative mechanism is divided into three main pathways (distal-alternating-enzymatic) according to the protonation steps of the N2 molecule. On the other hand, there are three main pathways in the associative mechanism, says N2.

- Abstract
- Introduction
- Computational Details
- Result and Discussion
- Catalytic Activity for NRR
- HER vs NRR
- Machine Learning Descriptors
- Conclusion
- References
Surprisingly, the free energy change in the last protonation step of *NH2 to *NH3 (* . refers to adsorbed species) is negative in the other electrocatalysts, which was usually considered a major obstacle (as a potentially determining step) for NRR in previous studies . GU (-eU) is the free energy related to the applied electrode potential, while U and e are the applied potential and the charge of the electron, respectively. For further understanding of NRR catalytic activity, we plotted free energy diagrams for the selected catalysts via distal/enzymatic pathways as shown in Figure 2.3.
In addition, the activation barrier for the first hydrogenation step was calculated, as this step shows a positive free energy change and is expected to affect the NRR. Charge analysis revealed that the charge transfer on the *N2 to *N2H (Figure 2.6a-c) adsorbate also has a high correlation with the free energy change of *N2 to *N2H. Furthermore, the free energy of the first hydrogenation step versus charge transfer on the *N2 adsorbate, as shown in Figure 2.7, shows the importance of charge transfer in N2 activation and subsequent protonation steps.
In order to better understand the last step of protonation, we analyzed the dependence of the free energy on the charge transfer (Figure 2.8). Gibbs free energy change for the first protonation step versus additional N2 charge on adsorbed N2H for (a) BA, (b) BP, and (c) BSb. d) and (e) Volcano plot for NRR using additional N2 charge on N2H adsorbate and *NNH adsorption energy (∆E (*NNH)) as descriptor. The Gibbs free energy change for the final protonation step as a function of the charge difference between *NH3 and.
We also trained a classification model (Figure 2.10) using kernel (radial basis. a) Free energy change for HER on the selected catalysts.

Abstract
Introduction
This process consumes almost 1-2% of global energy and releases 1% of total CO2 emissions.5 The electrochemical NRR has been intensively investigated,6, 7, which yields sustainability, flexibility and less environmental pollution. Many other aspects of electrochemical N2 reduction on defects such as modification of electronic structures and exploration of a wide range of materials need to be investigated, although some advances in defect engineering. The modification of the electronic structure is possible by the doping of metals in support through the interaction between support and TMs.
The aspects of active locations, supports and vacancies for NRR need to be further investigated as the NH3 yield is low and the HER still dominates. It is worth mentioning that mainly oxygen and nitrogen vacancies have been investigated in photocatalytic catalysts.29 Therefore, new vacancies are highly desirable to be investigated, for example Te, Se, S and C, by investigating NRR intermediate steps. We note that it may be possible for certain elements to be located in a hollow space between atoms (not in a lattice point).
Many descriptors such as smooth overlap of atomic positions (SOAP)31, Coulomb matrix32 and partial radial distribution function have been introduced.33 It is worth noting that some ML models have been presented for NRR and it is crucial to find more universal descriptors.34, 35 The above issue was a driving force for us to develop an ML method based on important intermediate steps involved in NRR. Some promising strategies for improving catalytic activity and a new mechanism of NRR are introduced, resulting in reducing the overpotential and suppressing HER by using different active sites (HS, IVD, TMs).
Computational Details
Here, we investigated 8 types of MBeners (Figure 3.1a) and 21 kinds of 2D materials (Figure 3.2b, Table 3.1) systematically by considering hollow sites (HS) and creating intrinsic vacancy defects (IVD) of three atoms (Te) respectively. See ya). In addition, we study 2D π-conjugated polymer (2DCP) as substrate surface for SACs on which various TMs are anchored while considering end-on and side-on modes of N2 adsorption (Figure 3.1c). The projected crystal orbital Hamiltonian population (pCOHP), which was used by the Lobster program41, 42, was used to determine the bonding and anti-bonding states between adsorbates and support.
The limiting potential is obtained by the step that has the most positive free energy change as follows: Ulimiting = ‒ΔGmax/e. E E E (XTe,Se,S) for defect formation was used where E[Vacancy] is the total energy with vacancy, E[total] is the total energy without vacancy, and µX is the chemical potential of X is (Te, Se, S). The solvent effect was not taken into account because many previous studies confirm that the solvent effect for NRR is approximately 0.1 eV.
Result and Discussion
- Activation of N 2 on Catalysts
- Machine Learning
- N 2 Reduction Mechanisms
- Potential Determining Step (PDS) in NRR
- NRR against HER
In addition, 2DCP-SACs are as good as graphene-based SACs for the activation of N2 through side-on mode. In Figure 3.2a-c, the π orbitals become non-degenerate by adsorption of an N2 molecule on an active center, which significantly shifts the orbitals of N2. On the other hand, the stretch in NN bond length is more evidence of the injection of electrons into antibonding π* orbitals of N2.
These results indicate that surface N2 activation (N-N bond length) is one of the main vital factors. It is worth noting that due to the strong activation of N2 on NbB, the length of the NN bond is significantly stretched in two initial steps. It was also shown that the excess of electrons on the *N-N adsorbate is proportional to the change in the free energy of *N2 to *N2H.
Among different mechanisms of NRR (Figure 3.7), the dissociative-associative and the enzymatic (associative) mechanisms are dominant for defective 2D materials/MBeners and 2DCP-SACs, respectively. The free energy changes of *N2 and *H are calculated in the active site of selected catalysts.

Conclusion
The binding energy of N2 is more negative than H2O for selected catalysts except TaSe2 and Nb@SAC, and it is expected that active sites will not be blocked by H2O molecules. Also, the flow of N2 builds up the gas pressure in the system, which can lead to a constant removal of water molecules from the surface. Interestingly, it can be said that the most effective TMs for NRR in terms of catalytic activity and selectivity are cases with both occupied and empty d-orbitals (starting elements of 3d, 4d and 5d blocks of the periodic table).
Combination of all the crucial features in selected catalysts (MBenes, defective 2D materials, 2DCP) showing low overpotential, high stability/selectivity and marked electrical conductivity (MBenes, 2DCP) will introduce new electrocatalysts. In the case of MBenes, the hollow sites introduce high potential catalysts for NRR (TaB with the highest NRR selectivity). By performing ML on datasets, new descriptors for NRR are presented with a combination of BOP and simple elementary features.
Linear trait-to-trait correlations show that N-N bond length is highly correlated with catalytic activity, indicating that N2 activation is critical for high catalytic performance. The HER overpotentials of the selected catalysts (TaB, NbTe2, NbB, HfTe2, MoB, MnB, HfSe2, TaSe2 and Nb@SAC) are more positive than the free energy change from N2 to *N2, which improves the Faraday efficiency .
Li, Q.; Liu, C.; Qiu, S.; Zhou, F.; He, L.; Zhang, X.; Sun, C., Investigation of iron borides as electrochemical catalysts for the nitrogen reduction reaction. Yang, X.; Shang, C.; Zhou, S.; Zhao, J., MBenes: Emerging 2D materials as efficient electrocatalysts for nitrogen reduction reaction. Lv, C.; Qian, Y.; Yan, C.; Ding, Y.; Liu, Y.; Chen, G.; Yu, G., Defect engineering metal-free polymeric carbon nitride electrocatalyst for effective nitrogen fixation under ambient conditions.
Yan, D.; Li, H.; Chen, C.; Zou, Y.; Wang, S., Defect engineering strategies for nitrogen reduction reactions under ambient conditions. Jiang, Z.; Wang, P.; Jiang, X.; Zhao, J., MBene (MnB): a new type of 2D metal ferromagnet with high Curie temperature. Liu, X.; Jiao, Y.; Zheng, Y.; Jaroniec, M.; Qiao, S.-Z., Building a picture of the electrocatalytic nitrogen reduction activity of transition metal single atom catalysts.
Zheng, S.; Li, S.; Mei, Z.; Hu, Z.; Chu, M.; Liu, J.; Chen, X.; Pan, F., Electrochemical nitrogen reduction reaction performance of single boron catalysts introduced by MXene substrates. Luo, Y.; Jiang, J., Realizing a non-strong-non-weak polarization electric field in single-atom catalysts sandwiched by boron nitride and graphene sheets for efficient nitrogen binding.
- Abstract
- Introduction
- Computational Details
- Results and Discussion
- Feature Engineering
- Classification by Deep Neural Network (DNN)
- Regression by Light Gradient Boosting Machine (LightGBM)
- Feature Importance and Correlation
- Adsorption of N 2 , N 2 H, NH 2 , NH 3 and H on SACs
- Stability and Free Energy Pathways
- Conclusion
- References
Based on previous studies, we use a three-key-step method to select active electrocatalysts for NRR.51-53. The adsorption energy of *N2 on an active site is considered to be more negative than -0.5 eV for efficient catalysts. Feature-feature correlation map (correlation values in %; EC: eligible catalyst). a) Feature-feature correlation map (correlation values in %; EC: eligible catalyst). b) The most important ranking features predicted by random forests (RF-blue columns) and mutual information (MI) methods. With this screening, more than half of the samples are filtered, which may control the overall screening rate for NRR.
For the final analysis, only samples with GNH3Desorbed0.8eV were considered as the most promising catalysts for NRR. Moreover, the configuration at the end shows more feasibility for NRR than the configuration at the side. Among the best catalysts, only HfB1C2 was shown to be more feasible for NRR via the lateral mechanism.
Furthermore, the end configuration appears to be more favorable for NRR than the lateral configuration (all data considered). The free energy change of the first hydrogenation step for Ru (0001), which has the lowest overpotential for NRR among the transition metals, is ~1 eV.13 but the PDS free energy change for B-doped SAC is much lower than 1 eV. DFT-calculated free energy diagrams for NRR via the distal mechanism for a) CrB3C and b) TcB3C1, sequential mechanism for c) HfB1C2, on B-doped SACs at zero and applied potential.
Our study demonstrates a great potential of machine learning methods for screening electrocatalysts for NRR and may motivate additional research to explore other highly efficient catalysts.
