Boosting the catalytic performance of metalloporphyrin-based covalent organic frameworks via coordination engineering for CO₂ and O₂ reduction

Abstract

Metalloporphyrin-based covalent organic frameworks (Por-COFs) are emerging as electrocatalysts; however, so far experiments have primarily focused on M–N₄-coordinated Por-COFs. A wealth of coordination modified systems with potential catalytic capability remains to be explored. Herein, we report a proof-of-concept computational study on the coordination engineering of Por-COFs as electrocatalysts for reducing CO₂ and O₂. Systematic density functional theory calculations were performed on 15 types of heteroatomic Por-COFs, featuring M–N_xO_yS_z (M = Fe, Co, and Ni; x + y + z = 4) centers. Calculations predicted that the Co–N₂O₂-Por-COF is an optimal CO₂-to-CO catalyst candidate (limiting potential ) and the Ni–N₂S₂-Por-COF is an optimal O₂-to-H₂O catalyst candidate (overpotential η^ORR = 0.46 V); both heteroatomic-Por-COFs display better catalytic activity and selectivity than their corresponding parent M–N₄-Por-COFs. Electronic structure analysis attributes the enhanced catalytic performance to the additional stabilization, endowed by the heteroatoms, of critical reaction intermediates. Furthermore, feature importance analysis based on machine learning models confirmed that the interplay between the central metal and the coordinated atoms is crucial for catalytic performance. This work predicts new Por-COFs as electrocatalysts for CO₂ and O₂ reduction and showcases the great potential of coordination regulation strategies in designing high-performance COF-based electrocatalysts.

---

1 Introduction

Covalent organic frameworks (COFs) are emerging electrocatalysts for energy conversion and storage, primarily due to their well-defined active sites, tunable coordination and topology, and accessible pore surfaces.^1–5 COFs are organized by organic knots and covalently connected linkers. Recently, the incorporation of metallo-porphyrin (Por) or phthalocyanine (Pc) knots into COFs has resulted in highly active, selective, and stable catalysts for electrochemical reduction of CO₂ and/or O₂.^6–28 In 2015, the Yaghi group developed the first cobalt porphyrin-based COF (i.e., COF-366-Co), which enables a 26-fold increase in electrochemical CO₂ reduction reaction (CO₂RR) activity compared to the molecular porphyrin catalyst.⁶ The Jiang group synthesized the CoPc-PDQ-COF, which enables a 32-fold increase in CO₂-to-CO activity compared to the molecular phthalocyanine catalyst.²⁵ In addition, the Co-TEPP-COF/C composite and FePc-BBL-COF displayed high electrocatalytic activity and selectivity for the reduction of O₂ to water.^26,27

Of note, the abovementioned Por/Pc-based COFs share a M–N₄-coordination centre. Coordination engineering, which alters the N₄-environment to heteroatoms, is an effective strategy to improve the catalytic properties of two-dimensional atomic dispersed transition metal catalysts in N-doped carbon, i.e. M–N–C single atom catalysts (SACs).^29–33 For example, Zhao et al. designed Ni-based carbon nanosheet SACs with NiN₃S sites and Song et al. designed Cu-based carbon support SACs with CuN₃O or CuCO₃ sites; all of these catalysts showed high catalytic activity and selectivity for CO production.^34,35 Li et al. synthesized an ortho-coordinated FeN₂O₂@graphene catalyst and achieved enhanced oxygen reduction reaction (ORR) performance in zinc–air batteries.³⁶ Wang et al. synthesized a carbon-supported Ni-based SAC with N₂O₂ coordination for high-efficiency O₂ reduction to H₂O₂.³⁷ Theoretically, the performance of graphene-based MS_xN_y (M = Fe, Co, Ni, and Cu) and FeXY_iN_3−i (X, Y = B, C, O, P, and S) SACs was investigated, leading to promising catalysts for electrochemical CO₂RR or water electrolysis, respectively.^38,39 However, to our knowledge, in experiments, coordination engineering has not been applied in Por/Pc-based COFs yet.

As a matter of fact, molecular heteroatomic porphyrins and metallo-heteroatomic porphyrins complexes have already been successfully synthesized.^40–49 These include singly-substituted cores, such as N₃O^41–43 and N₃S,^44–47 as well as doubly-substituted cores, such as N₂O₂,⁴¹ N₂S₂,^47,48 and N₂OS.⁴⁹ The embedded central metals include Fe, Co, Ni, Cu, Re, and Ru. In addition, H. Kim et al. have shown that Ni coordinated with 21-oxatetraphenylporphyrin (N₃O-TPP) becomes highly active toward electrochemical CO₂-to-CO reduction, unlike the barely active NiN₄–TPP complex.⁵⁰ These achievements make us believe that it is possible to synthesize heteroatomic-Por-COFs in the future. Accordingly, we anticipate that the incorporation of O and/or S atoms into the porphyrin unit will change the catalytic properties of the COFs, thus affording a series of new catalysts. Hence, a priori theoretical computation is important.

Herein, we present a proof-of-concept computational study on the coordination engineering of Por-COFs for catalysing the reduction of CO₂ and O₂, highlighting their importance in carbon neutrality and fuel cells. Through systematic periodic density functional theory calculations on M–N_xO_yS_z coordinated Por-COFs (Fig. 1), where M = Fe, Co, and Ni, we confirmed that coordination engineering is an effective strategy to boost the catalytic performance of Por-COFs. Calculations predict the Co–N₂O₂-Por-COF to be the optimal catalyst candidate for CO₂RR while the Ni–N₂O₂-Por-COF is optimal for ORR. The essence of enhanced catalytic performance was further investigated through electronic structure analysis and machine learning models. This study provides a fundamental mechanistic understanding of a new group of Por-COF electrocatalysts and identifies several candidates for CO₂RR/ORR, thus paving the way toward computation-guided design of potential electrocatalysts.

Fig. 1 (a) Unit cell of the M–N₄-Por-COF (M = Fe/Co/Ni) considered in this work and the constructed N_xO_yS_z coordination models. Reaction pathways for (b) CO₂RR-to-CO and (c) ORR-to-H₂O₂ (2e⁻)/H₂O (4e⁻).

2 Computational details

2.1 Calculation model setup

The metal-anchored Por-COF shown in Fig. 1a was used as the parent model for coordination engineering. It has been readily synthesized with 5,10,15,20-tetrakis(4-benzaldhyde)porphyrin and p-phenylenediamine (PPDA).^17,51 A periodic model of the Por-COF was used for calculations throughout this work. The unit cell of monolayer M–N_xO_yS_z-Por-COFs, containing 60 carbon atoms, 8 nitrogen atoms (or one or two to be substituted with oxygen and/or sulphur atoms), 36 hydrogen atoms, and 1 metal atom, was used. Furthermore, a 20 Å vacuum layer was placed on top of the surface to prevent mirror interactions.

2.2 Density functional theory (DFT) calculation methods

Spin polarized periodic DFT calculations were performed using the Vienna Ab initio Simulation Package (VASP).^52,53 Electron exchange and correction were described using the Perdew–Burke–Ernzerhof (PBE) functional within the generalized gradient approximation (GGA).^54,55 Grimmes's D3 scheme of dispersion correction was included to describe the van der Waals (vdW) interaction.⁵⁶ The plane wave basis was employed using the projector augmented wave (PAW) method with a cutoff energy of 520 eV.^57,58 The DFT+U calculations were performed and the corresponding U–J values are listed in Table S1 (ESI†).^59,60 Brillouin zone was sampled using 3 × 3 × 1 and 5 × 5 × 1 k-point grids for geometry optimization and electronic structure calculations, respectively. All atoms were allowed to relax during geometry optimization, and the convergence criteria for force and energy were set to be 10⁻² eV Å⁻¹ and 10⁻⁵ eV, respectively. A vacuum layer of 20 Å was placed between adjacent layers to prevent interaction. For M = Fe/Co/Ni, different spin states were tested and the discussed data correspond to the most stable structures (Tables S2–S4, ESI†). The implicit solvent model implemented in VASPsol with a dielectric constant of 78.4 was used for the aqueous solution.^61,62

2.3 Reaction free energy calculation

The computational hydrogen electrode (CHE) model developed by Nørskov et al.⁶³ was applied to obtain the thermodynamic Gibbs free energy change (ΔG) of all elementary reactions:

ΔG = ΔE + ΔE_ZPE − TΔS + ΔG_U + ΔG_pH (1)

where ΔE, ΔE_ZPE and ΔS are the difference in electronic energy, zero-point energy correction and entropy between the product and the reactant of each elementary reaction, respectively, and T is set as 298.15 K. E_ZPE and ΔS were obtained by calculating the vibrational frequencies for the adsorption intermediates (Tables S5 and S6, ESI†). ΔG_U and ΔG_pH represent the free energy corrections caused by the electrode potential (U) and the H⁺ concentration, respectively. Specifically, ΔG_U = −neU, where n is the number of transferred electrons and U is the applied potential; ΔG_pH = k_BT × ln 10 × pH, where k_B is the Boltzmann constant. The pH is set as 0 in this work for convenience, as has been done in most works.^18,38,39,64 The stability of the designed COFs under highly acidic conditions requires further experimental examination, but existing literature supports that certain M-Por-COFs are stable in acidic solutions.^65–67 In a few cases, energy values were computed at pH = 7 for comparison. It is important to note that the relative catalytic capability of the designed COF catalysts remains the same under different pH conditions.

The potential-determining step (PDS) is the step with the largest reaction Gibbs free energy value (ΔG_max) among all elementary steps. The limiting potential of CO₂RR is defined as U_L = −ΔG_max/e. The overpotential (η^ORR) of ORR is defined as η = U⁰ − U_L, where U⁰ is the equilibrium potential.

2.4 Stability evaluations

The thermodynamic and electrochemical stabilities of Por-COFs were investigated using binding energy (E_b), the difference between adsorption energy and cohesive energy (E_b − E_coh) and dissolution potential (U_diss). E_b measures the binding energy between the metal atom and the substrate,⁶⁸E_coh measures the cohesion tendency of isolated metal atom to form the bulk metal,^69,70U_diss measures the ability of active metal atoms to resist dissolution under electrochemical conditions;⁷¹ the related equations are as follows

E_b = E_Por-COF+M − E_Por-COF − E_single-M (2)

E_coh = E_bulk − n × E_single-M (3)

where E_COF+M, E_COF, E_single-M and E_bulk are the total energy of the Por-COF with M, the Por-COF without M, the single M atom and the total energy of metal atoms in the most stable bulk structure, respectively; and N_e are the standard dissolution potential of bulk metal and the number of electrons involved in dissolution, respectively. E_b < 0, E_b − E_coh < 0, and U_diss > 0 indicate that the catalyst has excellent thermodynamic and electrochemical stability. These criteria have been used to evaluate the stability of single atom catalysts in several works.^60,72,73

The dynamic stability of Co–N₂O₂, Ni–N₂S₂-, and Fe–N₂OS-Por-COFs was further tested through ab initio molecular dynamics (AIMD) simulations using the VASP package. A constant volume/temperature (NVT) ensemble and a Nose–Hoover thermostat were used to maintain the temperature at 300 K. The equation of motion is integrated using the Verlet algorithm method with a timestep of 2 fs, and the total simulation time is 10 ps for each system.

2.5 Machine learning (ML)

All machine learning (ML) predictions were conducted using an open-source Scikit-learn library within the Python3 environment.⁷⁴ The optimum parameter values for the three different ML algorithms used in this work, i.e., random forest regression (RFR),⁷⁵ gradient boosting regression (GBR),^76,77 and extreme gradient boosting regression (XGBoost),⁷⁸ are given in Table S7 (ESI†). Twenty widely recognized features relevant to active center properties were selected as input features for training ML models. All data sources were obtained either from existing databases or through single-step optimization calculations.⁷⁹ The input dataset was randomly divided into a training set and a test set in a 4 : 1 ratio. Given the constrained dataset size and potential errors stemming from the segmentation process, a fourfold cross-validation approach was implemented to ensure the model's generalizability. This approach entails the creation of four distinct training and test sets for ML model training and evaluation over four iterations. The optimal algorithm for and η^ORR prediction was determined by comparing the average values of root mean square error (RMSE) and coefficient of determination value (R² score) across four iterations; then, the best prediction on the test set was recorded for discussion.

3 Results and discussion

As shown in Fig. 1a, the metal-anchored Por-COF with PPDA as a linker was used as the parent model for coordination engineering. Three non-noble transition metal centers were considered: M = Fe, Co, and Ni. The four coordinated N atoms were partially substituted with either O or S atoms, leading to 5 different coordination models: N₃O, N₃S, ortho-N₂O₂, ortho-N₂S₂, and ortho-N₂OS. Taking together, a total of 15 types of M–N_xO_yS_z-Por-COFs were designed. The optimized structures (Fig. S1, ESI†) of the porphyrin macrocycles of M–N₃O- and M–N₂O₂-Por-COFs maintain the square-planar geometry, as observed in M–N₄-Por-COFs, whereas the substituted S atoms are positioned out of the plane due to their bulkier atom size. The observation of plane distortion caused by the bulkier S atom has been noted in molecular N₃S-, N₂S₂- and N₂OS-porphyrins,^40,46,49 as well as in M–N–C graphene-based SACs.^80,81

3.1 Stability and conductivity

The thermodynamic and electrochemical stabilities of heteroatomic Por-COFs were evaluated using binding energy (E_b), the difference between binding energy and cohesive energy (E_b − E_coh) and dissolution potential (U_diss). Calculated values (Fig. 2a–c) suggest that the N₃O-, N₃S and N₂O₂-coordinated Por-COFs, similar to the experimentally synthesized M–N₄-Por-COFs, meet all the stability criteria, i.e., E_b < 0, E_b − E_coh < 0 and U_diss > 0, meaning that the materials are able to avoid migration and aggregation of metal atoms under actual electrochemical work conditions. However, for the N₂S₂- and N₂OS-coordinated Por-COFs, the values of (E_b − E_coh) are slightly positive and U_diss values are slightly negative, which is likely attributed to the structure deformation introduced by S atoms. Hence, we further tested their dynamic stability using ab initio molecular dynamics (AIMD) simulations at 300 K. Throughout the 10 ps simulation, the systems maintain their integrity with the metal atoms sitting in the center position, confirming their structural stability (Fig. S2, ESI†).

Fig. 2 (a) The binding energy, (b) the difference between binding energy and cohesive energy, (c) the dissolution potential and (d) the band gap of Fe/Co/Ni–N_xO_yS_z-Por-COFs.

Then, the electrical conductivity of the designed materials was evaluated based on their band gaps (E_g). The E_g values of parent M–N₄-Por-COFs are 1.21, 1.39 and 1.45 eV for M = Fe, Co, and Ni, respectively (Fig. 2d and Fig. S3, ESI†). Such large gaps indicate their poor electron transfer ability, consistent with the limited current densities observed in experiments.¹⁷ Encouragingly, the calculated E_g of heteroatomic Por-COFs decreases greatly, suggesting an improved conductivity. Especially, the calculated band gap of Fe–N₃O-, Fe–N₃S-, Co–N₃S-, Co–N₂O₂-, Ni–N₂O₂- and Ni–N₂OS-Por-COFs vanishes, thus displaying conductor characteristics.

Accordingly, substituting the N-atom with O/S-atoms in the metal coordination of M–N₄-Por-COFs largely maintains stability and improves conductivity, serving as the foundation of further exploration of catalytic performance.

3.2 Catalytic performance

To investigate the effect of the coordination environment on the catalytic performance of Por-COFs, we studied the electrocatalytic reduction reaction of CO₂ and O₂ in sequence.

3.2.1 Reduction of CO₂ to CO. The conversion of CO₂ to CO is the dominant product of CO₂RR catalysed by Por-COFs,^{6–11,15,17–19,22,28,82} so it is the focus of this study. As illustrate in Fig. 1b, the pathway of converting CO₂ to CO catalysed by Por-COFs follows these steps: (1) adsorption of CO₂, (2) formation of carboxylic acid intermediate (*COOH) via the 1st proton-coupled electron transfer (PCET) step, (3) dehydration of *COOH intermediates to form *CO intermediates via the 2nd PCET step, and (4) desorption of CO to regenerate the catalyst.

The first step is the adsorption of CO₂. The absorption strength was slightly changed by substituting the N₄-coordination with N_xO_yS_z-coordination, with ΔE_ads-CO2 ranging from 0.02 to −0.34 eV (Table S8, ESI†). This is consistent with previous reports.⁶⁴ Meanwhile, depending on the coordination type the adsorbed shape of CO₂ on Por-COFs (Fig. S4 and S5, ESI†) can either be linear (for the case of N₄, N₃S, and N₂S₂) or bent (for the case of N₂O₂, N₃O, and N₂OS). For the bent case, the M–C_*CO2 bond is as short as 1.96 to 2.17 Å, along with a stretching of the C–O bond by 0.05 to 0.10 Å (ca. 4.3–8.6% elongation).

Then, the following catalytic steps for CO₂-to-CO conversion were computed and are presented in the free energy diagrams (Fig. S6–S8, ESI†). As calculated, the potential determining step (PDS) for all the target M–N_xO_yS_z-Por-COFs is the formation of the *COOH intermediate (Fig. 3a and Fig. S6, ESI†). Fig. 4a summarizes the corresponding theoretical limiting potentials U_L. In general, changing the coordination from N₄ to N_xO_yS_z shifts the U_L^CO2RR towards the positive direction, meaning an easier reduction and thus higher CO₂-to-CO catalytic activity. This is largely due to the stabilization of the *COOH intermediate, thus lowering the free energy of the PDS (ΔG_*→*COOH). In comparison, the U_L values of the Co-system are generally less negative than those of the Fe-/Ni-counterparts (Fig. 4a). Among all the designed Fe/Co/Ni–N_xO_yS_z-Por-COFs, the Co–N₂O₂-Por-COF (U_L = −0.58 V) is predicted to be the best CO₂-to-CO catalyst, followed by Fe–N₂OS (U_L = −0.64 V), Fe–N₂O₂ (U_L = −0.66 V) and Co–N₃O-Por-COFs (U_L = −0.67 V). The above results evidenced that modifying the M–N₄ coordination via introducing O or/and S atoms is an effective strategy to enhance the CO₂RR catalytic activity of Por-COFs.

Fig. 3 Free energy diagrams of (a) CO₂RR on Co–N₄-and Co–N₂O₂-Por-COFs and (b) ORR on Ni–N₄-and Ni–N₂S₂-Por-COFs. Optimized configurations of (c) *COOH, *CO, and (d) *OOH, *O, *OH intermediates. Color code: Co, purple; Ni, green; C, gray; N, blue; O, red; S, yellow; and H, white. For clarity, the tetraphenyl and p-phenylenediamine (PPDA) moieties of M–N_xO_yS_z-Por-COFs are not shown. Note: the corresponding free energy diagrams computed at pH = 7 are presented in Fig. S18 (ESI†) for comparison.

Fig. 4 (a) Limiting potential of CO₂RR-to-CO of Fe/Co/Ni–N_xO_yS_z-Por-COFs. (b) Limiting potential of CO₂RR-to-CO vs. limiting potential of HER. (c) Gibbs free energy changes (ΔG_{*OOH → *O}) of *O formation vs. Gibbs free energy changes (ΔG_{*OOH → HOOH}) of HOOH formation. (d) Overpotential of ORR-to-H₂O of Fe/Co/Ni–N_xO_yS_z-Por-COFs.

In addition, modifying the N₄ coordination also enhances the selectivity of CO₂RR over HER. Unlike the original three M–N₄-Por-COFs which exhibit a strong competition between CO₂RR and HER (Fig. 4b and Fig. S9, S10, ESI†), the majority of the designed M–N_xO_yS_z-Por-COFs display a distinct preference for CO₂RR. Exceptions are Ni–N₃O, Co–N₃O, Co–N₃S and Co–N₂S₂ systems, which a show slight preference for HER. Such undesired preference can be reversed in practice by adjusting the pH of the electrolyte or tuning the composition of the electrochemical double layer.^83–85

Although not the focus of this study, a great portion of spin distribution was observed on the porphyrin ligands, as shown in Co–N₂O₂–COOH intermediates (Fig S11, ESI†). This indicates the non-innocent role of the porphyrin ligands, which has been suggested by Liao et al.^86–89 This observation gives us a hint that modifying the porphyrin ligand could further tune the catalytic ability of the Por-COFs.

Based on the screening process considering intrinsic stability, catalytic activity, and selectivity, four CO₂-to-CO catalyst candidates are predicted: Co–N₂O₂, Fe–N₂OS, Fe–N₂O₂ and Co–N₃O-Por-COFs. Remarkably, all of them display performance comparable to available Por-COFs synthesized through experiments (Fig. S12, ESI†). It is worth noting that these newly designed Por-COFs may yield deep-reduction C1 products (CH₃OH and CH₄), especially when combined with photosensitizers, sacrificial electron donors, or supporting materials.^{22,82,87,89–91} These aspects merit further exploration, but we have limited this work to CO₂-to-CO reduction.

In brief, incorporating O/S atoms to modulate the coordination environment of active metals in Fe/Co/Ni-Por-COFs has been proven to be useful for CO₂-to-CO catalyst design. To explore the generality of this coordination modulation strategy, we now redirect our attention to another crucial electroreduction reaction, ORR.

3.2.2 Oxygen reduction reaction. Typically, the catalyzed ORR under acidic conditions can proceed through either a two-electron (2e⁻) pathway, resulting in the production of H₂O₂, or a four-electron (4e⁻) pathway, resulting in the production of H₂O. Fig. 1c shows the specific reaction pathways.

The adsorption of O₂ is the first step of ORR. In comparison, all the O/S coordinated Por-COFs show stronger adsorption of O₂ than the parent M–N₄-systems, with the exception of the Ni–N₂S₂-system. Observing the most stable O₂-adsorped configurations on the M–N_xO_yS_z-Por-COF (Fig. S13–S15, ESI†) shows that Fe-systems favour a side-on configuration (resembling the Griffiths-type) while Co/Ni-systems, except the Co–N₂OS case, favour an end-on configuration (resembling the Pauling-type). The side-on *O₂ structures exhibit a greater O–O bond stretch relative to the free O₂ (over 14% elongation for Fe-systems) than that of end-on *O₂ (roughly 10% elongation for Co/Ni-systems), indicating stronger activation of O₂ in the former (Fig. S16 and Table S9, ESI†). The calculated O₂ adsorption energy (ΔE_ads-O2) ranges from −0.18 to −1.58 eV, indicating chemisorption of O₂.

Since the 2e⁻ and 4e⁻ ORR pathways share the first two steps (* + O₂ → *O₂ and *O₂ → *OOH) and diverge at the *OOH intermediate, the selectivity is determined by the competition between the free energy difference of ΔG_*OOH→*HOOH and ΔG_*OOH→*O. Fig. 4c shows that all the considered catalysts favour the 4e⁻ ORR pathway over the 2e⁻ ORR pathway, except for the Ni–N₄-system shows similar preference, whose ΔG_*OOH→*O is −0.67 eV and ΔG_*OOH→*HOOH is −0.71 eV. Hence, we will focus on the 4e⁻ ORR pathway, which converts O₂ to H₂O, in the following discussion.

The effect of changing coordination from N₄ to N_xO_yS_z on 4e⁻ ORR activity is central-metal-dependent: the ORR activity is enhanced when M = Ni, however it is inhibited when M = Fe or Co (Fig. 4d). An ideal 4e⁻ ORR catalyst requires the ΔG of each elementary step to be ∼−1.23 eV at zero potential, but normally the differential adsorption/desorption behavior of the intermediates will disrupt this balance and results in overpotentials (η^ORR). Specifically, each catalyst has a different PDS for the 4e⁻ ORR (Fig. 3b and Fig. S17, ESI†): 1st step (* + O₂ → *OOH) for Ni–N₄, Co–N₄, and Co–N₂S₂ system; 2nd step (*OOH → *O) for the Ni–N₂S₂ system, and 4th step (*OH → *H₂O) for the rest. Although the parent M–N₄ catalysts have lower η^ORR values for Fe–N₄ (0.66 V) and Co–N₄ (0.43 V) than for Ni–N₄ (0.78 V), changing the coordinated environment to N_xO_yS_z only lowers the η^ORR of Ni-systems. These heteroatom-coordinated Ni-systems are Ni–N₃O (η^ORR = 0.74 V), Ni–N₂O₂ (0.72), Ni–N₂OS (0.63 V), Ni–N₃S (0.51 V) and Ni–N₂S₂ (0.46 V). The increase in η^ORR for heteroatom-coordinated Fe/Co-systems is attributed to the excessive strong adsorption of some intermediates (Fig. S19–S21, ESI†). Of note, the parent Co–N₄-catalysts exhibits the lowest η^ORR of 0.43 V, followed by the Ni–N₂S₂ (0.46) and Ni–N₃S (0.51) systems. All of them are promising ORR electrocatalyst candidates as compared to the commercial Pt (111) catalyst (η^ORR = 0.45 V).⁶³

3.3 Activity origin

The above calculations predict that, among the modified Por-COFs, the Co–N₂O₂-Por-COF is the best catalyst candidate for CO₂ reduction and the Ni–N₂S₂-Por-COF is the best candidate for O₂ reduction. To understand the reason behind the improved catalytic performance compared to parent Co–N₄/Ni–N₄-Por-COFs, electronic structure analyses were performed. It has been found that by changing the coordination from N₄ to N₂O₂/N₂S₂, the key intermediates associated with the PDSs were stabilized due to a better match between the orbitals of the central metal and the adsorbed species, thus improving catalytic activity.

In terms of comparison between Co–N₂O₂ and Co–N₄, *COOH serves as the key intermediate for CO₂-to-CO conversion (Fig. 3a). Changing the coordination from N₄ to N₂O₂ increases the orbital overlap between Co's 3d orbitals and the *COOH’ 2p orbitals, as depicted in Fig. 5a and b. The charge transferred from Co to the COOH-moiety (Fig. 5c and d) increases from 0.15 e⁻ (Co–N₄) to 0.32 e⁻ (Co–N₂O₂), and the respective Co−C_*COOH bond length decreases from 1.88 to 1.86 Å (Fig. 3c). Hence, the improved matching of critical orbitals leads to the stabilization of the *COOH intermediate, thus lowering the ΔG of step * + CO₂ → *COOH (Fig. 3a).

Fig. 5 Projected electronic densities of states (PDOS) and charge density difference (CDD) of (a)–(d) *COOH intermediates adsorbed on Co–N₄- and Co–N₂O₂-Por-COFs, and (e)–(h) *OOH intermediates adsorbed on Ni–N₄- and Ni–N₂S₂-Por-COFs. The cyan/yellow colors indicate the regions of electron loss/gain. Isosurfaces of charge density are set to 0.005 e Å ⁻³. For clarity, the tetraphenyl and the p-phenylenediamine (PPDA) moieties of M–N_xO_yS_z-Por-COFs are not shown.

Similarly, when comparing Ni–N₄ and Ni–N₂S₂ for the 4e⁻ ORR, the key intermediate *OOH was stabilized (Fig. 3b). A greater orbital overlap between Ni 3d orbitals and the 2p orbital of the adsorbed species was found as the coordination of Ni shifted from N₄ to N₂S₂ (Fig. 5e, f and Fig. S22, ESI†). The charge transferred from the substrate to intermediates increases from 0.28 e⁻ to 0.49 e⁻ for *OOH (Fig. 5g and h), from 0.33 e⁻ to 0.43 e⁻ for *O (Fig. S22, ESI†), and from 0.46 e⁻ to 0.57 e⁻ for *OH. The differential stabilization of these species collectively contributes to the adsorption–desorption equilibrium of reaction intermediates, thus decreasing the overpotential of the Ni–N₂S₂-Por-COF (Fig. 3b).

Overall, the increased catalytic performance of Co–N₂O₂-and Ni–N₂S₂-Por-COFs is attributed to the stabilization of critical intermediates via the introduction of heteroatoms into the coordination. Such enhanced effects due to heteroatoms have also been observed in multiple electrocatalysts.^92–94

3.4 Machine learning (ML)

Furthermore, we used ML to find the intrinsic characteristics that contribute to the differential CO₂RR or ORR performance of the above Por-COFs. As listed in Table 1, twenty widely recognized features were selected as input features for training the ML models. Numeric values of the features are present in Table S10 (ESI†). Since the size of the sample set is small, 18 in our case (14 in the training set and 4 in the test set), we tried three algorithms: the random forest regression (RFR), gradient boosting regression (GBR), and extreme gradient boosting regression (XGBoost) algorithms. Then, GBR was selected for predicting and XGBoost was selected for predicting η^ORR, as these models yield the lowest root mean square error (RMSE) and the highest coefficient of determination value (R² score) for the training set (Fig. 6 and Tables S11, 12, ESI†). A good linear correlation was achieved between the DFT-calculated and the ML-predicted values (Table S13, ESI†) of both (Fig. 6a) and η^ORR (Fig. 6b). The data of the test set are distributed near the diagonal line, indicating that the trained ML models are capable of predicting the catalytic activity of M-N_xO_yS_z-Por-COFs for CO₂RR and ORR, where M = Fe/Co/Ni.

Table 1 Features used for training the ML models

Category	Features	Abbreviation
Properties of central metal	Atomic number	N ^atom
	Number of valence electron	n e
	Number of d electron	n d
	Covalent radius	r ^atom
	van der Waals radius	r ^vdW
	Relative mass	m
	Pauling electronegativity	EN
	Electron affinity	EA
	First ionization energy	IE
Properties of coordinated atoms	Number of coordinated N atoms	n N
	Sum of valence electron count	∑ne
	Sum of p electron	∑np
	Sum of covalent radius	∑r^atom
	Sum of van der Waals radius	∑r^vdw
	Sum of Pauling electronegativity	∑EN
	The sum of electron affinity	∑EA
DFT calculated bond length	M–X1	d M–X1
	M–X2	d M–X2
	M–N1	d M–N1

---

Fig. 6 Comparison between DFT-calculated and ML-predicted values of (a) and (b) η^ORR. Feature importance for (c) in the GBR model and (d) η^ORR in the XGBoost model.

Then, feature importance analysis was performed based on the selected models to evaluate the significance of the features. For , the most important features (Fig. 6c) are the relative mass of central metals, m (33.53%); the M–X₂ bond length, d_M–X2 (28.26%); and the sum of electron affinity of coordination environments, ∑EA (12.80%). Other bond length features, including d_M–X1, d_M–N2, and d_M–N1 (each accounts for ∼3%), also have some importance. For η^ORR, the most crucial features (Fig. 6d) are the M–X₁ bond length, d_M–X1 (56.01%), and the atomic number of central metals, N^atom (27.47%); followed by the number of coordinated N atoms, n_N, (4.51%), the relative mass of central metals, m (4.42%), the sum of electron affinity of coordination environments, ∑EA (3.70%) and the sum of covalent radius of coordination environments, ∑r^atom (2.54%). The remaining features contribute insignificantly (Table S14, ESI†). The fact that the important features of CO₂RR and ORR are different is not unexpected, because their inherent mechanisms are different. As a matter of fact, one can see that no single feature overwhelmingly dominates the prediction. Hence the interplay between the central metal and the coordinated atoms is very important for determining the catalytic capability.

We further tested the applicability of the trained model in predicting CO₂RR or ORR catalytic performance of systems beyond the training set, i.e., central metal being Cu or Pt, and coordination atoms including C (refer to ESI† Note S1 for further details). There is a broad agreement on the catalytic order, which is inspiring; nevertheless, the match between ML-predicted and DFT-calculated values is not satisfactory (∼0.3 V error). We anticipate that increasing the sample size and type in the training set would improve the accuracy of the model, and this is currently underway in our group.

4 Conclusions

In summary, by adopting the strategy of substituting N-atoms with O/S-atoms in the metallo-porphyrin unit, our calculations have identified several types of N_xO_yS_x-coordinated metallo-Por-COFs as catalyst candidates for CO₂ or O₂ electroreduction. Since current efforts to enhance the catalytic performance of Por-COFs mainly focus on modifying the metal or the length or type of the linker, this study offers an alternative promising strategy: coordination engineering.

Changing the N₄-coordination to N_xO_yS_z-coordination (1) improved the conductivity of the M-Por-COFs (M= Fe/Co/Ni), while maintaining their stability; (2) enhanced the CO₂-to-CO reduction catalytic capability and selectivity against the HER in general; and (3) did not alter the preference for the 4e⁻ ORR in comparison to the 2e⁻ path, and increased the ORR activity for M = Ni, while hindering it for M = Fe/Co. We predict that the Co–N₂O₂-Por-COF is the best catalyst candidate for CO₂-to-CO reduction and the Ni–N₂S₂-Por-COF (η^ORR = 0.46 V) is the best candidate for ORR-to-H₂O reduction. The improved catalytic performance, as compared to their parent M–N₄-system, is attributed to a better orbital match between the central metal and the adsorbed species, leading to the stabilization of the critical intermediate of the potential determining step. Feature importance analysis based on machine leaning models showed that the relative mass of central metals and the bond length between metal and coordinated heteroatoms are the most important intrinsic characteristics for activity prediction.

This study showcases that the coordination engineering is an effective strategy for improving the catalytic performance of Por-based COFs. We envision that this strategy will have broader applications beyond Por-COFs. Although this study presents valuable findings, the mechanisms were solely based on the energetics of stationary points and neglected the kinetics. Further exploration of the detailed mechanisms of the best catalyst candidates is warranted and the impact of pH and explicit solvent molecules shall be taken into account. Besides, although the machine learning model provides information on the weight of important features, the predictability of the model is unsatisfactory. To improve the predictability and transferability of the model, it is necessary to expand the size and diversity of the dataset. Furthermore, the combination of changing the coordination environment and the linkers is also worth exploring.

Author contributions

Zhixin Ren: conceptualization, investigation, methodology, formal analysis, visualization, and writing – original draft. Ke Gong: formal analysis. Bo Zhao: visualization. Shi-Lu Chen: methodology and writing – review and editing. Jing Xie: conceptualization, visualization, supervision, resources, project administration, and writing – review and editing.

Conflicts of interest

The authors declare that they have no conflict of interest.

Acknowledgements

This work is supported by the National Natural Science Foundation of China (no. 22273004 and 22173007), the Beijing Natural Science Foundation (no. 2222028), the Teli Fellowship, the Innovation Foundation (no. 2021CX01026) from the Beijing Institute of Technology (BIT), China, and the BIT Research and Innovation Promoting Project (no. 2023YCXY043).

Notes and references

1. A. P. Côté, A. I. Benin, N. W. Ockwig, M. O'Keeffe, A. J. Matzger and O. M. Yaghi, Porous, crystalline, covalent organic frameworks, Science, 2005, 310, 1166–1170.
2. P. J. Waller, F. Gándara and O. M. Yaghi, Chemistry of Covalent Organic Frameworks, Acc. Chem. Res., 2015, 48, 3053–3063.
3. C. S. Diercks and O. M. Yaghi, The atom, the molecule, and the covalent organic framework, Science, 2017, 355, eaal1585.
4. J. Li, X. Jing, Q. Li, S. Li, X. Gao, X. Feng and B. Wang, Bulk COFs and COF nanosheets for electrochemical energy storage and conversion, Chem. Soc. Rev., 2020, 49, 3565–3604.
5. Z. Alsudairy, N. Brown, A. Campbell, A. Ambus, B. Brown, K. Smith-Petty and X. Li, Covalent organic frameworks in heterogeneous catalysis: recent advances and future perspective, Mater. Chem. Front., 2023, 7, 3298–3331.
6. S. Lin, C. S. Diercks, Y.-B. Zhang, N. Kornienko, E. M. Nichols, Y. Zhao, A. R. Paris, D. Kim, P. Yang, O. M. Yaghi and C. J. Chang, Covalent organic frameworks comprising cobalt porphyrins for catalytic CO₂ reduction in water, Science, 2015, 349, 1208–1213.
7. C. S. Diercks, S. Lin, N. Kornienko, E. A. Kapustin, E. M. Nichols, C. Zhu, Y. Zhao, C. J. Chang and O. M. Yaghi, Reticular Electronic Tuning of Porphyrin Active Sites in Covalent Organic Frameworks for Electrocatalytic Carbon Dioxide Reduction, J. Am. Chem. Soc., 2018, 140, 1116–1122.
8. P. L. Cheung, S. K. Lee and C. P. Kubiak, Facile Solvent-Free Synthesis of Thin Iron Porphyrin COFs on Carbon Cloth Electrodes for CO₂ Reduction, Chem. Mater., 2019, 31, 1908–1919.
9. Q. Wu, R.-K. Xie, M.-J. Mao, G.-L. Chai, J.-D. Yi, S.-S. Zhao, Y.-B. Huang and R. Cao, Integration of Strong Electron Transporter Tetrathiafulvalene into Metalloporphyrin-Based Covalent Organic Framework for Highly Efficient Electroreduction of CO₂, ACS Energy Lett., 2020, 5, 1005–1012.
10. H.-J. Zhu, M. Lu, Y.-R. Wang, S.-J. Yao, M. Zhang, Y.-H. Kan, J. Liu, Y. Chen, S.-L. Li and Y.-Q. Lan, Efficient electron transmission in covalent organic framework nanosheets for highly active electrocatalytic carbon dioxide reduction, Nat. Commun., 2020, 11, 497.
11. S. An, C. Lu, Q. Xu, C. Lian, C. Peng, J. Hu, X. Zhuang and H. Liu, Constructing Catalytic Crown Ether-Based Covalent Organic Frameworks for Electroreduction of CO₂, ACS Energy Lett., 2021, 6, 3496–3502.
12. M. Chen, H. Li, C. Liu, J. Liu, Y. Feng, A. G. H. Wee and B. Zhang, Porphyrin- and porphyrinoid-based covalent organic frameworks (COFs): From design, synthesis to applications, Coord. Chem. Rev., 2021, 435, 213778.
13. W. Ji, T.-X. Wang, X. Ding, S. Lei and B.-H. Han, Porphyrin- and phthalocyanine-based porous organic polymers: From synthesis to application, Coord. Chem. Rev., 2021, 439, 213875.
14. Z. Liang, H.-Y. Wang, H. Zheng, W. Zhang and R. Cao, Porphyrin-based frameworks for oxygen electrocatalysis and catalytic reduction of carbon dioxide, Chem. Soc. Rev., 2021, 50, 2540–2581.
15. Q. Wu, M.-J. Mao, Q.-J. Wu, J. Liang, Y.-B. Huang and R. Cao, Construction of Donor–Acceptor Heterojunctions in Covalent Organic Framework for Enhanced CO₂ Electroreduction, Small, 2021, 17, 2004933.
16. J.-Y. Yue, Y.-T. Wang, X. Wu, P. Yang, Y. Ma, X.-H. Liu and B. Tang, Two-dimensional porphyrin covalent organic frameworks with tunable catalytic active sites for the oxygen reduction reaction, Chem. Commun., 2021, 57, 12619–12622.
17. H. Dong, M. Lu, Y. Wang, H.-L. Tang, D. Wu, X. Sun and F.-M. Zhang, Covalently anchoring covalent organic framework on carbon nanotubes for highly efficient electrocatalytic CO₂ reduction, Appl. Catal., B, 2022, 303, 120897.
18. M. Fang, L. Xu, H. Zhang, Y. Zhu and W.-Y. Wong, Metalloporphyrin-Linked Mercurated Graphynes for Ultrastable CO₂ Electroreduction to CO with Nearly 100% Selectivity at a Current Density of 1.2 A cm⁻², J. Am. Chem. Soc., 2022, 144, 15143–15154.
19. T. He, C. Yang, Y. Chen, N. Huang, S. Duan, Z. Zhang, W. Hu and D. Jiang, Bottom-Up Interfacial Design of Covalent Organic Frameworks for Highly Efficient and Selective Electrocatalysis of CO₂, Adv. Mater., 2022, 34, 2205186.
20. S. Huang, K. Chen and T.-T. Li, Porphyrin and phthalocyanine based covalent organic frameworks for electrocatalysis, Coord. Chem. Rev., 2022, 464, 214563.
21. M. Liu, S. Liu, C.-X. Cui, Q. Miao, Y. He, X. Li, Q. Xu and G. Zeng, Construction of Catalytic Covalent Organic Frameworks with Redox-Active Sites for the Oxygen Reduction and the Oxygen Evolution Reaction, Angew. Chem., Int. Ed., 2022, 61, e202213522.
22. Y.-R. Wang, H.-M. Ding, X.-Y. Ma, M. Liu, Y.-L. Yang, Y. Chen, S.-L. Li and Y.-Q. Lan, Imparting CO₂ Electroreduction Auxiliary for Integrated Morphology Tuning and Performance Boosting in a Porphyrin-based Covalent Organic Framework, Angew. Chem., Int. Ed., 2022, 61, e202114648.
23. S. Bhunia, A. Peña-Duarte, H. Li, H. Li, M. F. Sanad, P. Saha, M. A. Addicoat, K. Sasaki, T. A. Strom, M. J. Yacamán, C. R. Cabrera, R. Seshadri, S. Bhattacharya, J.-L. Brédas and L. Echegoyen, [2,1,3]-Benzothiadiazole-Spaced Co-Porphyrin-Based Covalent Organic Frameworks for O₂ Reduction, ACS Nano, 2023, 17, 3492–3505.
24. Y. Yuan, K.-T. Bang, R. Wang and Y. Kim, Macrocycle-Based Covalent Organic Frameworks, Adv. Mater., 2023, 35, 2210952.
25. N. Huang, K. H. Lee, Y. Yue, X. Xu, S. Irle, Q. Jiang and D. Jiang, A Stable and Conductive Metallophthalocyanine Framework for Electrocatalytic Carbon Dioxide Reduction in Water, Angew. Chem., Int. Ed., 2020, 59, 16587–16593.
26. G. Lu, H. Yang, Y. Zhu, T. Huggins, Z. J. Ren, Z. Liu and W. Zhang, Synthesis of a conjugated porous Co(II) porphyrinylene–ethynylene framework through alkyne metathesis and its catalytic activity study, J. Mater. Chem. A, 2015, 3, 4954–4959.
27. Z. Zhang, W. Wang, X. Wang, L. Zhang, C. Cheng and X. Liu, Ladder-type π-conjugated metallophthalocyanine covalent organic frameworks with boosted oxygen reduction reaction activity and durability for zinc-air batteries, Chem. Eng. J., 2022, 435, 133872.
28. X. Zhang, Y. Yuan, H. Li, Q. Wu, H. Zhu, Y. Dong, Q. Wu, Y. Huang and R. Cao, Viologen linker as a strong electron-transfer mediator in the covalent organic framework to enhance electrocatalytic CO₂ reduction, Mater. Chem. Front., 2023, 7, 2661–2670.
29. X. Li, H. Rong, J. Zhang, D. Wang and Y. Li, Modulating the local coordination environment of single-atom catalysts for enhanced catalytic performance, Nano Res., 2020, 13, 1842–1855.
30. Y. Zhu, J. Sokolowski, X. Song, Y. He, Y. Mei and G. Wu, Engineering Local Coordination Environments of Atomically Dispersed and Heteroatom-Coordinated Single Metal Site Electrocatalysts for Clean Energy-Conversion, Adv. Energy Mater., 2020, 10, 1902844.
31. W. Xu, H. Tang, H. Gu, H. Xi, P. Wu, B. Liang, Q. Liu and W. Chen, Research progress of asymmetrically coordinated single-atom catalysts for electrocatalytic reactions, J. Mater. Chem. A, 2022, 10, 14732–14746.
32. K. Liu, P. Chen, Z. Sun, W. Chen, Q. Zhou and X. Gao, The atomic interface effect of single atom catalysts for electrochemical hydrogen peroxide production, Nano Res., 2023, 16, 10724–10741.
33. X. Xie, H. Peng, G. Ma, Z. Lei and Y. Xu, Recent progress in heteroatom doping to modulate the coordination environment of M–N–C catalysts for the oxygen reduction reaction, Mater. Chem. Front., 2023, 7, 2595–2619.
34. X. Zhao, S. Huang, Z. Chen, C. Lu, S. Han, C. Ke, J. Zhu, J. Zhang, D. Tranca and X. Zhuang, Carbon nanosheets supporting Ni–N₃S single-atom sites for efficient electrocatalytic CO₂ reduction, Carbon, 2021, 178, 488–496.
35. P. Song, B. Hu, D. Zhao, J. Fu, X. Su, W. Feng, K. Yu, S. Liu, J. Zhang and C. Chen, Modulating the Asymmetric Atomic Interface of Copper Single Atoms for Efficient CO₂ Electroreduction, ACS Nano, 2023, 17, 4619–4628.
36. Y. Li, Y. Ding, B. Zhang, Y. Huang, H. Qi, P. Das, L. Zhang, X. Wang, Z.-S. Wu and X. Bao, N,O symmetric double coordination of an unsaturated Fe single-atom confined within a graphene framework for extraordinarily boosting oxygen reduction in Zn–air batteries, Energy Environ. Sci., 2023, 16, 2629–2636.
37. Y. Wang, R. Shi, L. Shang, G. I. N. Waterhouse, J. Zhao, Q. Zhang, L. Gu and T. Zhang, High-Efficiency Oxygen Reduction to Hydrogen Peroxide Catalyzed by Nickel Single-Atom Catalysts with Tetradentate N₂O₂ Coordination in a Three-Phase Flow Cell, Angew. Chem., Int. Ed., 2020, 59, 13057–13062.
38. P. Hou, Y. Huang, F. Ma, X. Wei, R. Du, G. Zhu, J. Zhang and M. Wang, S and N coordinated single-atom catalysts for electrochemical CO₂ reduction with superior activity and selectivity, Appl. Surf. Sci., 2023, 619, 156747.
39. S. Wang, B. Huang, Y. Dai and W. Wei, Tuning the Coordination Microenvironment of Central Fe Active Site to Boost Water Electrolysis and Oxygen Reduction Activity, Small, 2023, 19, 2205111.
40. T. Chatterjee, V. S. Shetti, R. Sharma and M. Ravikanth, Heteroatom-Containing Porphyrin Analogues, Chem. Rev., 2017, 117, 3254–3328.
41. P. J. Chmielewski, L. Latos-Grażyński, M. M. Olmstead and A. L. Balch, Nickel Complexes of 21-Oxaporphyrin and 21, 23-Dioxaporphyrin, Chem. – Eur. J., 1997, 3, 268–278.
42. M. Pawlicki and L. Latos-Grazyński, Iron complexes of 5,10,15,20-tetraphenyl-21-oxaporphyrin, Inorg. Chem., 2002, 41, 5866–5873.
43. S. Stute, L. Götzke, D. Meyer, M. L. Merroun, P. Rapta, O. Kataeva, W. Seichter, K. Gloe, L. Dunsch and K. Gloe, Molecular Structure, UV/Vis Spectra, and Cyclic Voltammograms of Mn(II), Co(II), and Zn(II) 5,10,15,20-Tetraphenyl-21-oxaporphyrins, Inorg. Chem., 2013, 52, 1515–1524.
44. J. Lisowski, L. Latos-Grazynski and L. Szterenberg, Nuclear magnetic resonance study of the molecular and electronic structure of nickel(II) tetraphenyl-21-thiaporphyrins, Inorg. Chem., 1992, 31, 1933–1940.
45. L. Latos-Grazynski, J. Lisowski, P. Chmielewski, M. Grzeszczuk, M. M. Olmstead and A. L. Balch, Palladium complexes of 21-thiaporphyrin: syntheses and characterization, Inorg. Chem., 1994, 33, 192–197.
46. I. Gupta and M. Ravikanth, Synthesis of meso-furyl porphyrins with N4, N3S, N2S2 and N3O porphyrin cores, Tetrahedron, 2003, 59, 6131–6139.
47. T. Kaur, A. Ghosh, P. Rajakannu and M. Ravikanth, Synthesis and crystal structure of the rhenium(I) tricarbonyl complex of 5,10,15,20-tetra-p-tolyl-21,23-dithiaporphyrin, Inorg. Chem., 2014, 53, 2355–2357.
48. T. Kaur, W.-Z. Lee and M. Ravikanth, Rhenium(I) Tricarbonyl Complexes of meso-Tetraaryl-21,23-diheteroporphyrins, Inorg. Chem., 2016, 55, 5305–5311.
49. S. Punidha, N. Agarwal, R. Burai and M. Ravikanth, Synthesis of N₃S, N₃O, N₂S₂, N₂O₂, N₂SO and N₂OS Porphyrins with One meso-Unsubstituted Carbon, Eur. J. Org. Chem., 2004, 2223–2230.
50. H. Kim, D. Shin, W. Yang, D. H. Won, H.-S. Oh, M. W. Chung, D. Jeong, S. H. Kim, K. H. Chae, J. Y. Ryu, J. Lee, S. J. Cho, J. Seo, H. Kim and C. H. Choi, Identification of Single-Atom Ni Site Active toward Electrochemical CO₂ Conversion to CO, J. Am. Chem. Soc., 2021, 143, 925–933.
51. Y. Meng, Y. Luo, J.-L. Shi, H. Ding, X. Lang, W. Chen, A. Zheng, J. Sun and C. Wang, 2D and 3D Porphyrinic Covalent Organic Frameworks: The Influence of Dimensionality on Functionality, Angew. Chem., Int. Ed., 2020, 59, 3624–3629.
52. G. Kresse and J. Hafner, Ab initio molecular dynamics for open-shell transition metals, Phys. Rev. B: Condens. Matter Mater. Phys., 1993, 48, 13115–13118.
53. G. Kresse and J. Furthmüller, Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set, Phys. Rev. B: Condens. Matter Mater. Phys., 1996, 54, 11169–11186.
54. J. P. Perdew, K. Burke and M. Ernzerhof, Generalized Gradient Approximation Made Simple, Phys. Rev. Lett., 1996, 77, 3865–3868.
55. S. Grimme, J. Antony, S. Ehrlich and H. Krieg, A consistent and accurate ab initio parametrization of density functional dispersion correction (DFT-D) for the 94 elements H-Pu, J. Chem. Phys., 2010, 132, 154104.
56. S. Grimme, S. Ehrlich and L. Goerigk, Effect of the damping function in dispersion corrected density functional theory, J. Comput. Chem., 2011, 32, 1456–1465.
57. P. E. Blöchl, Projector augmented-wave method, Phys. Rev. B: Condens. Matter Mater. Phys., 1994, 50, 17953–17979.
58. G. Kresse and D. Joubert, From ultrasoft pseudopotentials to the projector augmented-wave method, Phys. Rev. B: Condens. Matter Mater. Phys., 1999, 59, 1758–1775.
59. S. L. Dudarev, G. A. Botton, S. Y. Savrasov, C. J. Humphreys and A. P. Sutton, Electron-energy-loss spectra and the structural stability of nickel oxide: An LSDA + U study, Phys. Rev. B: Condens. Matter Mater. Phys., 1998, 57, 1505–1509.
60. H. Xu, D. Cheng, D. Cao and X. C. Zeng, A universal principle for a rational design of single-atom electrocatalysts, Nat. Catal., 2018, 1, 339–348.
61. K. Mathew, R. Sundararaman, K. Letchworth-Weaver, T. A. Arias and R. G. Hennig, Implicit solvation model for density-functional study of nanocrystal surfaces and reaction pathways, J. Chem. Phys., 2014, 140, 084106.
62. K. Mathew, V. S. C. Kolluru, S. Mula, S. N. Steinmann and R. G. Hennig, Implicit self-consistent electrolyte model in plane-wave density-functional theory, J. Chem. Phys., 2019, 151, 234101.
63. J. K. Nørskov, J. Rossmeisl, A. Logadottir, L. Lindqvist, J. R. Kitchin, T. Bligaard and H. Jónsson, Origin of the Overpotential for Oxygen Reduction at a Fuel-Cell Cathode, J. Phys. Chem. B, 2004, 108, 17886–17892.
64. S. Cao, S. Wei, X. Wei, S. Zhou, H. Chen, Y. Hu, Z. Wang, S. Liu, W. Guo and X. Lu, Can N, S Cocoordination Promote Single Atom Catalyst Performance in CO₂RR? Fe–N₂S₂ Porphyrin versus Fe–N₄ Porphyrin, Small, 2021, 17, 2100949.
65. A. Nagai, X. Chen, X. Feng, X. Ding, Z. Guo and D. Jiang, A Squaraine-Linked Mesoporous Covalent Organic Framework, Angew. Chem., Int. Ed., 2013, 52, 3770–3774.
66. C. Liu, H. Li, F. Liu, J. Chen, Z. Yu, Z. Yuan, C. Wang, H. Zheng, G. Henkelman, L. Wei and Y. Chen, Intrinsic Activity of Metal Centers in Metal-Nitrogen-Carbon Single-Atom Catalysts for Hydrogen Peroxide Synthesis, J. Am. Chem. Soc., 2020, 142, 21861–21871.
67. J.-D. Yi, R. Xu, G.-L. Chai, T. Zhang, K. Zang, B. Nan, H. Lin, Y.-L. Liang, J. Lv, J. Luo, R. Si, Y.-B. Huang and R. Cao, Cobalt single-atoms anchored on porphyrinic triazine-based frameworks as bifunctional electrocatalysts for oxygen reduction and hydrogen evolution reactions, J. Mater. Chem. A, 2019, 7, 1252.
68. S. Back, J. Lim, N. Kim, Y. Kim and Y. Jung, Single-atom catalysts for CO₂ electroreduction with significant activity and selectivity improvements, Chem. Sci., 2017, 8, 1090–1096.
69. Y. Su, L. Zhang, Y. Wang, J. Liu, V. Muravev, K. Alexopoulos, I. A. W. Filot, D. G. Vlachos and E. J. M. Hensen, Stability of heterogeneous single-atom catalysts: a scaling law mapping thermodynamics to kinetics, npj Comput. Mater., 2020, 6, 144.
70. M. Ha, D. Y. Kim, M. Umer, V. Gladkikh, C. W. Myung and K. S. Kim, Tuning metal single atoms embedded in N_xC_y moieties toward high-performance electrocatalysis, Energy Environ. Sci., 2021, 14, 3455–3468.
71. J. Greeley and J. K. Nørskov, Electrochemical dissolution of surface alloys in acids: Thermodynamic trends from first-principles calculations, Electrochim. Acta, 2007, 52, 5829–5836.
72. C. Fang, J. Zhou, L. Zhang, W. Wan, Y. Ding and X. Sun, Synergy of dual-atom catalysts deviated from the scaling relationship for oxygen evolution reaction, Nat. Commun., 2023, 14, 4449.
73. X. Guo, S. Zhang, L. Kou, C. Yam, T. Frauenheim, Z. Chen and S. Huang, Data-driven pursuit of electrochemically stable 2D materials with basal plane activity toward oxygen electrocatalysis, Energy Environ. Sci., 2023, 16, 5003–5018.
74. F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot and É. Duchesnay, Scikit-learn: Machine Learning in Python, J. Mach. Learn. Res., 2011, 12, 2825–2830.
75. L. Breiman, Random Forests, Mach. Learn., 2001, 45, 5–32.
76. J. H. Friedman, Greedy function approximation: A gradient boosting machine, Ann. Stat., 2001, 29, 1189–1232.
77. J. H. Friedman, Stochastic gradient boosting, Comput. Stat. Data Anal., 2002, 38, 367–378.
78. T. G. Dietterich, Approximate Statistical Tests for Comparing Supervised Classification Learning Algorithms, Neural Comput., 1998, 10, 1895–1923.
79. H. Mai, T. C. Le, D. Chen, D. A. Winkler and R. A. Caruso, Machine Learning for Electrocatalyst and Photocatalyst Design and Discovery, Chem. Rev., 2022, 122, 13478–13515.
80. Z. Jin, D. Jiao, Y. Dong, L. Liu, J. Fan, M. Gong, X. Ma, Y. Wang, W. Zhang, L. Zhang, Z. G. Yu, D. Voiry, W. Zheng and X. Cui, Boosting Electrocatalytic Carbon Dioxide Reduction via Self-Relaxation of Asymmetric Coordination in Fe-based Single Atom Catalyst, Angew. Chem., Int. Ed., 2023, e202318246.
81. H. Jia, A. Nandy, M. Liu and H. J. Kulik, Modeling the roles of rigidity and dopants in single-atom methane-to-methanol catalysts, J. Mater. Chem. A, 2022, 10, 6193–6203.
82. Z. Ren, B. Zhao and J. Xie, Designing N-Confused Metalloporphyrin-Based Covalent Organic Frameworks for Enhanced Electrocatalytic Carbon Dioxide Reduction, Small, 2023, 19, 2301818.
83. A. S. Hall, Y. Yoon, A. Wuttig and Y. Surendranath, Mesostructure-Induced Selectivity in CO₂ Reduction Catalysis, J. Am. Chem. Soc., 2015, 137, 14834–14837.
84. X.-Q. Li, G.-Y. Duan, J.-W. Chen, L.-J. Han, S.-J. Zhang and B.-H. Xu, Regulating electrochemical CO₂RR selectivity at industrial current densities by structuring copper@poly(ionic liquid) interface, Appl. Catal., B, 2021, 297, 120471.
85. S. Banerjee, C. S. Gerke and V. S. Thoi, Guiding CO₂RR Selectivity by Compositional Tuning in the Electrochemical Double Layer, Acc. Chem. Res., 2022, 55, 504–515.
86. Y.-Q. Zhang, J.-Y. Chen, P. E. M. Siegbahn and R.-Z. Liao, Harnessing Noninnocent Porphyrin Ligand to Circumvent Fe-Hydride Formation in the Selective Fe-Catalyzed CO₂ Reduction in Aqueous Solution, ACS Catal., 2020, 10, 6332–6345.
87. L.-L. Shi, M. Li, B. You and R.-Z. Liao, Theoretical Study on the Electro-Reduction of Carbon Dioxide to Methanol Catalyzed by Cobalt Phthalocyanine, Inorg. Chem., 2022, 61, 16549–16564.
88. Y.-C. Cao, L.-L. Shi, M. Li, B. You and R.-Z. Liao, Deciphering the Selectivity of the Electrochemical CO₂ Reduction to CO by a Cobalt Porphyrin Catalyst in Neutral Aqueous Solution: Insights from DFT Calculations, ChemistryOpen, 2023, 12, e202200254.
89. J.-Y. Chen, M. Li and R.-Z. Liao, Mechanistic Insights into Photochemical CO₂ Reduction to CH₄ by a Molecular Iron−Porphyrin Catalyst, Inorg. Chem., 2023, 62, 9400–9417.
90. H. Rao, L. C. Schmidt, J. Bonin and M. Robert, Visible-light-driven methane formation from CO₂ with a molecular iron catalyst, Nature, 2017, 548, 74–77.
91. Y. Wu, Z. Jiang, X. Lu, Y. Liang and H. Wang, Domino electroreduction of CO₂ to methanol on a molecular catalyst, Nature, 2019, 575, 639–642.
92. X. Li, H. Zhang, Q. Hu, W. Zhou, J. Shao, X. Jiang, C. Feng, H. Yang and C. He, Unlocking the Transition of Electrochemical Water Oxidation Mechanism Induced by Heteroatom Doping, Angew. Chem., Int. Ed., 2023, 62, e202309732.
93. C. Feng, M. Lv, J. Shao, H. Wu, W. Zhou, S. Qi, C. Deng, X. Chai, H. Yang, Q. Hu and C. He, Lattice Strain Engineering of Ni₂P Enables Efficient Catalytic Hydrazine Oxidation-Assisted Hydrogen Production, Adv. Mater., 2023, 35, 2305598.
94. Q. Hu, K. Gao, X. Wang, H. Zheng, J. Cao, L. Mi, Q. Huo, H. Yang, J. Liu and C. He, Subnanometric Ru clusters with upshifted D band center improve performance for alkaline hydrogen evolution reaction, Nat. Commun., 2022, 13, 3958.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d3qm01315d

---