Computational Study of Electrochemical CO₂ Reduction on 2D Graphitic Carbon Nitride Supported Single‐Atom (Al and P) Catalysts (SACs)

Abstract

To mitigate the adverse effects of CO₂ emissions, CO₂ electroreduction to small organic products is a preferable solution and potential catalysts include the single‐atom catalyst (SAC) which comprises individual atoms dispersed on 2D materials. Here, we used aluminium and phosphorus as the active sites for CO₂ electroreductions by embedding them on the 2D graphitic carbon nitride (g‐C₃N₄) nano‐surface. The resulting M‐C₃N₄ (M=Al and P) SACs were computationally studied for the CO₂ electroreduction using density functional theory (DFT) and ab‐initio molecular dynamics (AIMD) simulations. Computations showed that CO₂ can be adsorbed to the active sites in forms of a frustrated Lewis pair (Al/N or P/N) or single atom Al or P. The adsorbed CO₂ can be converted to various intermediates by gaining proton and electron (H⁺+e⁻) pairs, a process simulated as electroreduction. While both SACs prefer to produce HCOOH with low potential determining steps (PDSs) and small overpotential values of 0.25 V and 0.08 V for Al‐C₃N₄ and P‐C₃N₄ respectively, to produce CH₄, P‐C₃N₄ exhibits a lower potential barrier of 0.9 eV than Al‐C₃N₄ (1.07~1.17 eV).

DFT computational study showed that CO₂ can be electrochemically reduced to organic molecules including formic acid and methane, assisted by single‐atom catalysts by embedding Al or P on the monolayer of 2D graphitic carbon nitride. Further, we identified the potential determining steps, and discussed the selectivity and overpotentials in the production of C1 species from the reduction of CO₂.

Introduction

Human activities such as the combustion of fossil fuels (coal, oil, and natural gas), industrial processes, and deforestation accelerate the CO₂ emission into the atmosphere, which has resulted in severe environmental challenges such as global warming, ocean acidification, and melting glaciers.[ ¹ , ² , ³ , ⁴ ] There are two general strategies to reach the goal of zero‐emission of CO₂, i. e., capturing CO₂ and sequestering it deep in the ground or preferably converting it to value‐added commodity chemicals. The latter is advantageous both from a scientific as well as an environmental perspective, and various technologies for converting CO₂ to valuable chemicals such as formic acid (HCOOH), methanol (CH₃OH), and methane (CH₄) have been developed.[ ⁵ , ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ ] Among the viable technologies so far, the electrochemical reduction of CO₂ is a promising way for the common and cost‐effective process that enables a sustainable energy economy and chemical industry.[ ¹ , ⁵ , ⁹ ] However, due to the high overpotential, poor selectivity, low efficiency of Faraday, and the competition of hydrogen evolution reduction (HER),[ ¹² , ¹³ , ¹⁴ , ¹⁵ ] it remains challenging to facilitate the CO₂ reduction reaction (CO₂RR).[ ⁵ , ¹⁶ , ¹⁷ ] In addition, the conversion of CO₂ to small multicarbons requires lot of energy as CO₂ is a linear, thermally stable, nonpolar, and inert gas molecule.[ ¹⁸ , ¹⁹ , ²⁰ ] Obviously, catalysts are the core for CO₂RR and it is essential to discover novel catalysts for the electrocatalytic CO₂ activation and conversion processes which not only are efficient and selective, but preferably use earth‐abundant elements.

Currently, transition metals, especially noble metals including Au, Ag, Pt, and Pd based catalysts remain the most widely used ones for the CO₂ conversion with high selectivity.[ ²¹ , ²² , ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ ] Among the three types of catalysts (homogeneous, heterogeneous, and biological catalysts), heterogeneous catalysts are the most popular one utilized in industry since catalysts and reactants/products can be easily separated. But homogeneous catalysts are generally more selective and cost effective. A growing interest is to combine homogeneous and heterogeneous catalysts and develop low cost, highly active and selective catalysts. In this regard, single atom catalysts (SACs) have become a new frontier in catalysis due to their unique properties and potentials for enhanced catalytic performance and are expected to bridge the homogeneous and heterogeneous catalyst properties.[ ⁵ , ⁸ , ²⁵ , ²⁹ , ³⁰ , ³¹ , ³² ] SACs comprise atoms of metal species dispersed on solid supports such as metal oxides and carbonaceous materials and can mimic homogeneous catalysts where the atomically dispersed metal active centers serve as the active sites, providing similar reactivity and product selectivity. Thus, SACs are characterized with their excellent atomic economy and catalytic activity compared with traditionally supported nanoparticles. The concept of SAC was proposed and promoted in 2011 by Zhang et al., who demonstrated that a catalyst with isolated Pt atoms anchored to the surfaces of iron oxide nanocrystallites (i. e., Pt₁/FeO_x) was much superior in CO oxidation than its nanoparticle counterpart.[ ³³ , ³⁴ ] Sun et al. showed that the SAC Pt/graphene can be conveniently made through atomic layer deposition. ^[35] So far, numerous SACs have been designed and employed using two‐dimensional (2D) materials as supports which include graphene,[ ⁵ , ⁹ , ³⁶ , ³⁷ , ³⁸ , ³⁹ ] graphitic carbon nitride (g‐C₃N₄),[ ¹⁶ , ⁴⁰ , ⁴¹ , ⁴² , ⁴³ , ⁴⁴ ] molybdenum disulfide (MoS₂),[ ¹⁷ , ⁴⁵ , ⁴⁶ , ⁴⁷ ] and hexagonal boron nitride (h‐BN).[ ⁴⁸ , ⁴⁹ , ⁵⁰ , ⁵¹ ]

For the reduction of CO₂, both metal‐ and nonmetal‐embedded or doped SAC materials have demonstrated excellent catalytic and semiconducting properties.[ ⁸ , ⁹ , ³² , ⁵² ] For instances, computational study showed that CO₂ can be converted to various multicarbon products like C1 (CH₃OH and CH₄) and C2 (CH₂CH₂ and CH₃CH₂OH) on single and double boron atoms doped bismuthine, ^[53] or to CH₄ on transition metal nitrogen‐doped graphene. ^[5] Liu et al. evidenced that an oxygen atom embedded in RuN₄ doped graphene (RuN₄/O‐Gra) has better catalytic performance and better stability for the O₂ and CO₂ reduction reactions than RuN₄ doped graphene (RuN₄‐Gra). ^[36] Notably, recent work indicated that one C3 product propane can be generated from CO, which itself can be generated from CO₂, on a Fe‐embedded hexagonal boron nitride (hBN) nanosheet due to the very high electron accumulation on the hBN surface when metal atoms are embedded. ^[54] Xing et al. achieved the CO₂ electroreduction to hydrocarbon products by doping the single atoms of transition metals on a boron carbide monolayer theoretically. ^[55] SACs in the form of Co‐S_xN_4−x (x=0, 1, 2, 3) were synthesized by Pei et al. to facilitate the CO₂ electroreduction to CO. ^[56] Further, Jiang's group explored CO₂RR catalyzed by Fe and Ni single‐atom pairs embedded on MOF‐derived N‐doped carbon‐based materials ^[57] and a range of metal atoms on N‐doped carbon materials (abbreviated as M‐N‐C, where M=Fe, Co, Ni and Cu). ^[58]

One of the allotropes in 2D carbon‐based nanomaterials is graphitic carbon nitride (g‐C₃N₄), which consists of sp² hybridized C and N atoms and contains the structure of van der Waals layers. ^[9] As a metal‐free polymeric semiconductor, g‐C₃N₄ has been employed in many electrocatalytic reactions including CO₂RR due to a nitrogen‐rich carbon site and its band gap (~2.7 eV).[ ⁶ , ⁹ , ⁵⁹ ] The Chen group demonstrated that CO₂ can be captured on the phosphorus‐doped g‐C₃N₄ in photocatalytic reduction, and the optical absorption and the photoreactions of CO₂ and H₂O are enhanced due to the shifting of the band structure and the charge carriers, respectively. ^[60] In addition, Jiao et al. performed DFT computations and found that the electrocatalytic CO₂ reduction and the production rate of C2 species are better on Cu‐embedded g‐C₃N₄ than on Cu‐embedded N‐doped graphene because of its molecular scaffold facilitating coordination. ^[42] While extensive studies have been performed to explore the catalytic performance of embedded single metal atoms on the 2D graphitic carbon or nitrogen‐doped carbon nanosheet, further computational analyses are expected to provide detailed information of CO₂RR at the atomic and electronic levels and predict catalytic properties of new materials that have not yet been synthesized.

In this work, we intended to compare the electrochemical reaction mechanisms of CO₂RR catalyzed by two SACs by embedding aluminium (Al) and phosphorus (P) on the monolayer of g‐C₃N₄ by DFT simulations. As the non‐toxic and the most abundant metallic element on the earth, Al has a high charge density and a low electronegativity, making it a good Lewis acid for activating CO₂ and promoting the reduction by hydrogenation. On the other hand, P acts as a strong electron donor. It introduces Lewis basic sites, which are electron‐rich regions that can effectively interact with CO₂ molecules and facilitate the electron transfer to CO₂ during reduction reactions. While embedding these single atoms on g‐C₃N₄, we also identified frustrated Lewis pairs (FLP) such as N/Al or N/P on these 2D materials because recent research shows that surface FLPs can play a significant catalytic role for small molecule activation and hydrogenations.[ ³⁹ , ⁶¹ , ⁶² , ⁶³ , ⁶⁴ , ⁶⁵ , ⁶⁶ , ⁶⁷ , ⁶⁸ ] FLPs are characterized with the combination of a Lewis acid and a Lewis base whose steric hindrance obstructs the formation of a typical covalent bond between them.[ ⁶¹ , ⁶² , ⁶³ , ⁶⁴ , ⁶⁵ , ⁶⁶ , ⁶⁷ , ⁶⁹ ] FLPs of Al‐embedded nitrogen‐doped graphene have shown catalytic activity for the hydrogenation of CO₂. ^[39] In metal‐free catalyst systems, FLPs have been employed for the CO₂ activation and reduction processes. ^[68] This paper started from the CO₂ adsorption on these SAC surfaces. The adsorption occurs with either the single atoms Al and P of SACs or paired Al/N and P/N, i. e., FLPs. Afterwards, we elucidated the CO₂RR mechanisms on M‐C₃N₄ (M=Al and P) in details. One of the simplest mechanisms in the CO₂ reduction is the production of HCOOH as the following steps (“*” denotes the catalytic surface)

(1)

(2)

(3)

However, we discussed all likely reduction steps by transferring electrons and protons in the process of CO₂RR using DFT simulations. Finally, based on the computations, we identified the potential determining steps, and discussed the selectivity and overpotentials in the production of C1 species from the reduction of CO₂.

Computational Methods

All simulation models were built on graphitic carbon nitride (g‐C₃N₄) with the embedding of single atom aluminium or phosphorus. The lattice parameters of the optimized g‐C₃N₄ surface were 7.403×7.403×20.000 Å, with the angle between the first two vectors being 120°. Spin‐polarized DFT methods were used to carry out the electronic structure calculations by Vienna Ab Initio Simulation Package (VASP).[ ⁷⁰ , ⁷¹ ] The projector augmented wave (PAW) potential described the interactions between valance electrons and ions.[ ⁷² , ⁷³ ] The Perdew‐Burke‐Ernzerhof (PBE) functional within the generalized gradient approximation (GGA) was used to evaluate exchange‐correlation interactions.[ ⁷⁴ , ⁷⁵ ] The kinetic energy cutoff was set to 450 eV with plane‐wave bases in all calculations. A 5×5×1 Monkhorst‐Pack mesh of k‐point grids was adopted for all surfaces of the unit cells. The total energy of the surfaces was converted to 10⁻⁶ eV by applying the Gaussian smearing method with a width of 0.05 eV. All atomic positions of these surfaces were allowed to relax until the maximum forces on atoms were below 0.01 eV/Å. A vacuum space, 20 Å, was added along the Z‐direction to avoid interactions between periodic units during the structural relaxation. Electron transfers were analyzed in different models of surfaces by the Bader charge population analysis. ^[76] To explain the thermal stability of 2D materials, ab initio molecular dynamics (AIMD) simulations were performed at the temperatures of 300 and 500 K using the canonical ensemble with a Nose‐Hoover thermostat for 10 ps.[ ⁷⁷ , ⁷⁸ ] The binding energy of the single atom on this surface was defined as

$${\displaystyle {\Delta \mathrm{E}}_b=E_{M-C_3{N}_4}-E_{C_3{N}_4}-E_M}$$ (4)

where $E_{M-C_3{N}_4}$ is the total DFT energy of an embedded single atom on the surface of g‐C₃N₄, $E_{C_3{N}_4}$ is the energy of the monolayer of g‐C₃N₄, and $E_M$ is the energy of the embedded atom. The adsorption energy of all intermediates was computed as

$${\displaystyle {\Delta \mathrm{E}}_{ads}=E_{Total}-E_{M-C_3{N}_4}-E_{Adsorbate}}$$ (5)

where $E_{Total}$ is the total energy of adsorbed species on M‐C₃N₄, and $E_{M-C_3{N}_4}$ and $E_{Adsorbate}$ are the corresponding energies of M‐C₃N₄ and free adsorbate, respectively.

To compute the Gibbs free energy change (ΔG) in each reduction step of CO₂RR, the computational hydrogen electrode (CHE) method was used.[ ⁷⁹ , ⁸⁰ ] According to this model, the free energy of the proton‐electron pair (H⁺+e⁻) can be computed to the chemical potential of H₂ under the standard condition at 0 V vs. standard hydrogen electrode (SHE). In brief, ΔG for each reduction step can be computed as

$${\displaystyle \Delta G=\Delta E+{\Delta }_{ZPE}-\mathrm{T}\Delta S+{\Delta G}_U+{\Delta G}_{pH}}$$ (6)

where $\Delta E$ represents the electronic energy change obtained from DFT simulations, ${\Delta }_{ZPE}$ is the zero‐point energy change, and $\Delta S$ refers to the entropy change at the temperature (T) of 298.15 K. ${\Delta G}_U$ in the equation can be written as ${\Delta G}_U=-neU$ where $U$ means the applied electrode voltage, and $ne$ provides the amount of charges during each reduction step. ${\Delta G}_{pH}$ is equal to $K_{B}T+ln10+pH$ , where $K_B$ is the Boltzmann constant and $pH$ is set to be zero value. The entropy and vibrational frequencies of gas molecules are taken from the NIST database. The rate‐limiting potential ($U_L$ ) and the overpotential (η) were calculated with the following equations.

$${\displaystyle U_L=-\frac{\Delta G}{e}}$$ (7)

$${\displaystyle \eta =U_{eq}-U_L}$$ (8)

Results and Discussion

Geometric Structures and Thermal Stability of Al‐C₃N₄ and P‐C₃N₄

The geometrical structures of the monolayer g‐C₃N₄ (001) surface embedded with the single metal atom Al or non‐metal atom P in the cavity were obtained by fully relaxed optimizations (Figure 1). Both embedded atoms are stabilized in the six‐fold cavity close to a corner between two edges and bonded to maintain either a nonplanar (for Al) or a planar (for P) structure with two adjacent nitrogen atoms at the distances of 2.089 or 1.791 Å, respectively. In the nonplanar configuration of Al‐C₃N₄, the single atom Al protrudes from its surface by 1.1 Å, and the nonplanar form is more stable by 0.41 eV than the metastable planar form. More structural data are shown in Figure S1 (Supplementary Information). The bond distances of N−Al and N−P are close to previously reported values.[ ³⁹ , ⁸¹ , ⁸² ] One of the important properties in evaluating a catalyst is its thermal stability. For Al‐C₃N₄ and P‐C₃N₄, the single atom binding energy was calculated as −3.10 and −2.46 eV, respectively, suggesting that both 2D materials have good thermodynamic stability. Figures 2a1 and 2b1 show the charge density difference (CDD) plots of these 2D surfaces. It is interesting to note that the charge accumulation (yellow) and depletion (cyan) mainly concentrate on the embedded atoms and the two nearest N atoms, which are bonded to Al and P. Due to the significant electronegativity difference between Al/P and N, the charge accumulation and depletion indicate the electron density flow from Al/P to the adjacent N atoms. This was further confirmed by the Bader charge analyses which show significant charge transfer from Al and P to the substrate g‐C₃N₄ by 1.66 and 1.12 e, respectively. In other word, both Al and P atoms carry significant positive charges. Based on the DFT computations, we also obtained the electron spin density, electron‐localized function (ELF) plots, the total density of state (DOS), and projected density of state (PDOS). The electron spin densities (Figures 2a2 and 2b2) were for both surfaces of Al‐C₃N₄ and P‐C₃N₄ by DFT computations. They exhibited nonmagnetic behavior because there is no localized electron spin density around the Al and P atoms. ELF plots reveal the interactions and the characteristics of the bond types for the 2D surfaces with high (up to 1) and low (down to 0) ELF values. While the high ELF (close to 1) region in red represents the localized electrons by making an electron pair or a covalent bond, the value 0.5 indicates the delocalized region in green, and ELF=0 reflects the low charge density in blue. Clearly, the ELF plots (Figures 2a3 and 2b3) indicated that the electron densities on the Al and P atoms are strongly localized as either the covalent bonds between N and the SAC atoms or the lone pairs. Al embedded SAC showed stronger such localizations than P‐C₃N₄. The spin‐polarized total density of state (DOS) of these 2D materials were shown in Figures 3a1–b1 and 3a2–b2 respectively. Further, the partial DOS (PDOS) graphs in Figure S2 revealed overlaps between the 3p orbitals of Al or P and the 2p orbitals of the neighboring N atoms, indicating strong interactions in these SACs. By considering Figures 3a2 and 3b2 for the participation of 3p orbitals in the interactions between the single atom Al or P and g‐C₃N₄, Al‐C₃N₄ showed the sharper peaks of 3p_y and 3p_z around the Fermi level (E_F) than 3p_x. On the other hand, there were only 3p_z peaks near the E_F on the P‐C₃N₄ surface to interact with the 3p orbitals of C and N atoms. These DOS and PDOS plots of the surfaces exhibited the symmetric peaks both up and down spins, indicating the total electron spin density approximately zero. To verify the thermal stability of these 2D SACs, we conducted AIMD simulations with 2×2 supercells at temperatures of 300 and 500 K for 10 ps. Figure 4 showed the plots of the temperature (a1–b1), energy (a2–b2), and distance (a3–b3) vs time of the AIMD simulations of these nanomaterials at 300 K and results at 500 K were shown in the Supplementary Information (Figure S3). During the 10 ps period at 300 K and 500 K, there are few significant fluctuations in these supercells according to these plots except for the surface of Al‐C₃N₄ at 300 K, where Al atom shifts around the defect by bonding to another two N atoms in this sixfold cavity of g‐C₃N₄.

Figure 1.

Optimized structures of (a) Al‐C₃N₄ and (b) P‐C₃N₄, with C in black, N in blue, and Al or P in pink or orange, respectively.

Figure 2.

Charge density difference (a1/b1), electron spin density (a2/b2), and ELF (a3/b3) plots of (a) Al‐C₃N₄ and (b) P‐C₃N₄. The yellow and blue colors in a1 and b1 represent the accumulation and depletion of electron densities, respectively, and the isosurface values of the CDD plots (a1 and b1) are 0.005 e/ų. The isosurface value of the electron spin density for Al‐C₃N₄ (a2) and P‐C₃N₄ (b2) is 5×10⁻⁷.

Figure 3.

(a1/b1) Plots of the total density of states and (a2/b2) the projected density of states for individual orbitals of embedded atoms in Al‐C₃N₄ and P‐C₃N₄.

Figure 4.

Molecular dynamics simulations for the variation of (a) temperature (K), (b) energy (eV), and (c) distance (Å) along the time (ps) at 300 K on (a1–a3) Al‐C₃N₄ and (b1–b3) P‐C₃N₄ respectively.

CO₂ Adsorption and Activation on Al‐C₃N₄ and P‐C₃N₄ Surfaces

CO₂ is a linear and stable molecule, and bending CO₂ requires considerable energy input. On the other hand, CO₂ ⁻ is of a bent geometry. Thus, the activation of CO₂ is often ignited by the electron transfer to it. Figure 5 showed the three likely conformational structures with CO₂ adsorbed on Al‐C₃N₄. The optimal values of the bond lengths and the angles of the adsorbed CO₂ molecule in these configurations were listed in Table 1. For the adsorption of CO₂ on Al‐C₃N₄, one or two oxygen atoms of CO₂ are bonded with Al with CO bonds stretched (Table 1). Importantly, CO₂ turns into a non‐linear system by bending its ∠OCO angle to 125.8°, 131.2°, and 110.9° for SA1, SA2, and SA3, respectively. The net CO₂ adsorption energies in these structures are −0.26, −0.47, and −0.41 eV, respectively, but it should be noted that the bending would cost energy depending on the angle. ^[83] By adding the deformation energy cost, the interaction energy (ΔE _int , listed in Table 1) between M‐C₃N₄ and CO₂ may be more insightful than the binding energy which is the sum of the interaction and deformation energy components. The high interaction energies indicate strong chemical bonding and electron transfers. Thus, Al‐C₃N₄ prefers to adsorb CO₂ chemically, and the configuration SA3 exhibits the strongest interaction compared to the other two (SA1 and SA2). In all these three configurations, the oxygen atom(s) of CO₂ interact with the positively charged Lewis acidic Al. For the adsorption on P‐C₃N₄, however, we were able to obtain only two optimized structures, SP1 and SP2. This may result from the conflicting properties of P on P‐C₃N₄. Population analyses in the above already show that it carries significant amount of positive charge. But P is a Lewis base which is supposed to interact with the positively charged C atom of CO₂. This explains why the binding energy in SP2 is positive and SP2 is essentially a metastable (intermediate) state. In the configuration SP1, CO₂ is away from the surface by 3.50 Å with a negligible adsorption (0.06 eV) and retains a linear geometry. Thus, SP1 is simply a van der Waals (or pre‐adsorption) structure and is a pre‐state of SP2, where CO₂ is obviously activated with strong interaction energy. In the subsequent step, we simulated the reduction mechanisms on P‐C₃N₄ starting from the SP1 structure. Notably, SA1 exhibits the cooperativity between the interactions of Al and N with CO₂, a characteristic of FLPs, as both Al and N are typical Lewis acid and base, respectively. While the CDD plots were shown in Figure S4, the amounts of charge transferred from Al‐C₃N₄ to CO₂ in the configurations SA1, SA2, and SA3 are 0.49, 1.21, and 1.37 e, respectively. Since SP1 is a van der Waals state, the charge transfer is negligible. But in SP2, there is an electron transfer of 0.40 e from P‐C₃N₄ to CO₂, in accord with the bent structure of CO₂ therein. The interaction between CO₂ molecule and the surfaces can also be elucidated by PDOS plots (Figure S5).

Figure 5.

Optimized structures of Al‐C₃N₄ and P‐C₃N₄ after adsorption CO₂ onto site, SA1, SA2, SA3, SP1 and SP2.

Table 1.

Adsorption (${\Delta \mathrm{E}}_{ads}$ , in eV), interaction energy (${\Delta \mathrm{E}}_{int}$ , in eV), and relative energy (eV), charge transfer (Δq, in e), bond distances (R, in Å) and angles (∠, in degree) of CO₂ on M‐C₃N₄ surfaces (M=Al, P).

SAC	Struct	R(C−O)	∠OCO	R(SAC−O)	${\Delta \mathrm{E}}_{ads}$	${\Delta \mathrm{E}}_{int}$	Δq	Relative energy (eV)
Al‐C3N4	SA1	1.330/1.212	125.8	1.752	−0.26	−4.03	−0.49	0.21
	SA2	1.362/1.206	131.2	1.828	−0.47	−3.16	−1.21	0
	SA3	1.334/1.321	110.9	1.819/1.835	−0.41	−6.74	−1.37	0.06
P‐C3N4	SP1	1.176/1.178	177.8	3.663/3.637	−0.06	−0.07	−0.03	0.37
	SP2	1.368/1.208	124.7	1.725	1.08	−2.72	−0.40	0

CO₂ Reduction on Al‐C₃N₄ and P‐C₃N₄ Surfaces

With the adsorption and activation of CO₂ on Al‐C₃N₄ and P‐C₃N₄, we continued to explore the production of C1 molecules including HCOOH, CO, H₂CO, CH₃OH, and CH₄ in electrochemical processes with combined electron and proton transfers. The most stable configurations in all reduction processes were obtained by DFT computations and shown in Figures S6 and S7. Based on the literature and our computations, we illustrated the likely pathways for the electrocatalytic CO₂RR in Scheme 1, and the corresponding energy profiles were shown in Figure 6, where the reaction coordinate (x‐axis) represents the successive transfers of the electron and proton pairs to produce the intermediates.

Scheme 1.

Possible reaction pathways for CO₂RR on (a) Al‐C₃N₄ (b) P‐C₃N₄.

Figure 6.

Free energy diagram of CO₂ reduction pathways to C1 products on Al‐C₃N₄ starting from the surfaces (a1) SA1, (a2) SA2, (a3) SA3, and on P‐C₃N₄ by (b) SP1.

Reduction mechanism on Al‐SAC: The first electrochemical reduction step of CO₂RR on Al‐C₃N₄ is the hydrogenation of the adsorbed CO₂ to three distinct intermediates including *OCHO*, *COOH, and *O*COH, where “*” highlights the bonding site(s) to the surface. The free energy change for the formation of the first intermediate *OCHO* on Al‐C₃N₄ is −1.28 eV which is the lowest value by comparing with other intermediates, suggesting that *OCHO* is the preferred intermediate state. The second hydrogenation step on Al‐C₃N₄ leads to *OCH₂O*, *OCHOH, *OC*HOH, *OC*, and *OC through the first intermediates by either forging CH bonds or releasing a water molecule. The possible evolutions of these adsorbed species to produce free HCOOH and CO can be described with the following equations (also see the energy profiles in Figure S8:

(9)

(10)

(11)

(12)

While the rate‐limiting steps to produce HCOOH and CO through SA1 on Al‐C₃N₄ require to overcome the potential of 1.58 eV, for SA2 the reaction potentials are 0.5 eV and 0.67 eV for HCOOH and CO, respectively (see Figure S8). We note that although it shows a spontaneous reaction to form *OCHO* on SA3, the generation of the next intermediate *OCHO*H from *OCHO* on Al‐C₃N₄ requires more than 2 eV (Figure S8a). Hence, there is no plausible pathway to yield formic acid through SA3.

The production of CH₃OH, CH₄, and H₂CO from the second intermediates on this Al‐SAC is also likely. There are two pathways for the next round of hydrogenation step from *OC*HOH to either *OC*H by releasing one H₂O molecule or *OCH₂OH. While *OC*H eventually can be converted to *OC*H₂ on this Al‐SAC surface, the pathway from *OC*HOH to *OC*H₂ requires a reaction potential of 0.45 eV (Figure 6a1). Alternatively, the generation of *OCH₂OH from *OC*HOH is a spontaneous process. The intermediate *OCHOH similarly shows a spontaneous reduction to *OCH₂OH by a pair of H⁺ and e⁻ on Al‐C₃N₄ (Figure 6a2). Along another reaction pathway (Figure 6a3), *OCHO* turns to *OCH₂O* nearly spontaneously by gaining a pair of proton and electron, and the reduction of *OCH₂O* to *OCH₂O*H experiences a reaction potential barrier of 0.66 eV. Based on the energy profiles in Figures 6a1–a2, we found two pathways to yield H₂CO from *OC*H₂ and *OCH₂ as

(13)

(14)

The continuing fourth electroreduction on Al‐C₃N₄ leads *OCH₂OH to *O by producing CH₃OH with the ΔG free energy potential 0.80 eV. This step can be summarized as

(15)

The fifth hydrogenation step on Al‐C₃N₄ after Eq. 15 generates *OCH₃ from *OCH₂ and *OC*H₂ spontaneously by releasing energy. However, there would be a high potential (1.12 eV) for the formation of *OCH₃ through *OCH₂O*H and *OCH₃‐*OH. This step is rate‐determining if the whole CO₂RR starts from the SA3 configuration.

The production of CH₄ comes from the intermediate *OCH₃ which is electrochemically reduced as

(16)

with a low potential barrier 0.32 eV on Al‐C₃N₄.

Reduction mechanism on P‐SAC: The initial step of CO₂RR on P‐C₃N₄ results in the formation of only *OCHO and *OCOH. We note that the potential barrier is very high, i. e., 2.26 eV to reach the intermediate *OCOH in the first reduction step. Hence, it is prohibitive to reduce CO₂ through *OCOH. On P‐C₃N₄, the addition of an electron and proton pair to the first intermediate *OCHO leads to *HCOOH, followed by either the departure of a formic acid molecule from P‐C₃N₄ as

(17)

or further electrochemical reductions (Figure 6b). The above process leading to the yield of formic acid experiences a potential barrier of 0.33 eV on P‐C₃N₄.

The next reduction step on this surface is the formation of *OCH₂OH from *HCOOH, which experiences negligible energy change (0.03 eV). For the CO₂RR on P‐C₃N₄, the rate‐limiting step is observed in the fourth reduction step, namely from *OCH₂OH to *OCH₂ and one water molecule whose potential reaches 0.9 eV, followed by the generation of H₂CO. However, an alternative hydrogenation step results in *OCH₃ spontaneously, followed by the yield of methanol and methane in the sixth reduction step as Eq. 18 with the potentials of 0.27 eV and none, respectively.

(18)

We note that both Al‐C₃N₄ and P‐C₃N₄ reduce the intermediate *OCH₃ to CH₃OH with the same mechanism, but P‐C₃N₄ experiences a lower energy barrier. Further, P‐C₃N₄ can reduce *OCH₃ to methane with no potential barrier. During the electroreduction processes, we also obtained the adsorbed intermediate *O, which is bonded directly with Al or P atoms with the distance of 1.629 or 1.491 Å, respectively. *O can be successively reduced to *OH and then *OH₂ with a 1.07 eV potential height (see Figure 6a1‐a3). Similar mechanisms apply to the processes on P‐C₃N_4, though the formation of *OH and the release of H₂O require potentials of 0.42 eV and 0.18 eV, respectively (see Figure 6b). According to these reduction mechanisms, the productions of CO, HCOOH, H₂CO, CH₃OH, and CH₄ are favorable.

At last, we calculated the desorption free energy costs of the products, HCOOH, CO, H₂CO, CH₃OH, and CH₄, from the M‐C₃N₄ (M=Al and P) SACs, which were compiled in Table 2. These values indicate that all products are possible to be generated during the CO₂RR with low potential barriers (<1 eV) of the desorption energy except CH₃OH by SA3 on Al‐SAC. Also, the desorption of CH₄ is more preferred than CH₃OH due to the lower desorption potentials. Although the major products of Al‐C₃N₄ are CO and HCOOH, methane shows the highest spontaneous desorption energy potential as the major product on P‐C₃N₄.

Table 2.

Calculated desorption free energy cost (eV) for M‐C₃N₄ (M=Al and P) SACs.

SACs	Struct	CO	HCOOH	H2CO	CH3OH	CH4
Al‐C3N4	SA1	−1.95	0.29	0.49	0.80	0.32
	SA2	−0.58	0.10	0.65	0.80	0.32
	SA3	/	/	/	/	0.32
P‐C3N4	SP1	/	−0.11	−0.72	0.27	−1.48

Selectivity of Al‐C₃N₄ or P‐C₃N₄ in Electrochemical Reductions from CO₂ to CO, HCOOH, H₂CO, CH₃OH, and CH₄

Table 3 compares the rate‐limiting potentials and overpotentials for the CO₂ electroreduction to C1 products of formic acid (SA1, SA2, and SP1) and methane (SA3) on both Al‐C₃N₄ and P‐C₃N₄. The overpotentials of all other products are shown on Table S1. These values are based on the calculations of the maximum free energy changes after the protonation and electron transfers along the lowest energy paths. If the CO₂RR starts from the conformation SA1, the maximum free energy change in the rate‐limiting step is 1.58 eV to produce all these small organics. However, the maximum potential values starting from SA2 are less than 1 eV for the formation of CO and HCOOH on Al‐C₃N₄, suggesting the preferable production of these two species. From SA3, only CH₄ is likely.

Table 3.

Limiting potentials (U_L), potential determining step (PDS), and overpotentials (η) of electrochemical CO₂ reduction on M‐C₃N₄ (M=Al and P).

System (products)	PDS	UL (V)	Ueq (V)	η (V)
SA1 (HCOOH)	*+CO2+H++e−→*O*COH	−1.58	−0.250	1.33
SA2 (HCOOH)	*COOH+H++e−→*OCHOH	−0.5	−0.250	0.25
SP1 (HCOOH)	*+CO2+H++e−→*OCHO	−0.33	−0.250	0.08
SA3 (CH4)	*OCH3‐*OH+H++e−→*OCH3+H2O	−1.12	0.169	1.29

Unlike Al‐C₃N₄ which has multiple pathways from CO₂ to yield C1 products, P‐C₃N₄ shows only one mechanism to reduce CO₂ (Figure 6b). Along the pathway leading to HCOOH on P‐C₃N₄, the maximum energy potential is 0.33 eV. To produce H₂CO, CH₃OH, or CH₄, the rate‐limiting potentials are the same (0.9 eV), as all these products are generated from the intermediate state *OCH₃ by further protonation and electron transfers. Thermodynamically, however, CH₄ is the most preferable, followed by CH₃OH and H₂CO. Comparing the production of CH₄ by both SACs, the key step is Eq. 16. There is a negative potential energy change (−1.48 eV) on P‐C₃N₄ while the energy change is positive (0.32 eV) on Al‐C₃N₄. Thus, P‐C₃N₄ is a better catalyst for the electrochemical reduction of CO₂ to CH₄.

The overpotential plays a crucial role in determining the performance of CO₂RR on SACs. Lower overpotential values correspond to better catalysts in the reduction of CO₂ to C1 products. Table 3 summarizes the overpotentials, which are calculated with Eq. 8. In these calculations, we used the equilibrium potentials (U_eq) for different products, i. e., CO= −0.106 V, HCOOH=−0.250 V, H₂CO=−0.070 V, CH₃OH= 0.016 V, and CH₄= 0.169 V.[ ⁸⁴ , ⁸⁵ ] Previous research reported low overpotentials for the production of HCOOH and CO on Cu, Pd, Au, and so on.[ ²⁶ , ²⁸ , ⁸⁶ , ⁸⁷ , ⁸⁸ , ⁸⁹ , ⁹⁰ , ⁹¹ ] Among them, the catalysts with Cu, Bi, and Co (II) complexes show the overpotential values 0.25–0.45 V, 0.65 V, and 0.62 V, respectively, for yielding HCOOH.[ ⁸⁶ , ⁹² , ⁹³ ] Comparing these data with our simulated mechanisms for the production of formic acid, while the η value for Al‐SAC starting from SA2 is 0.25 V, it is 0.08 V for P‐SAC from the SP1 conformation. Hence, the overpotentials on M‐C₃N₄ (M=Al and P) SACs are even lower during the reduction from CO₂ to HCOOH. To generate CO, the potential value presents as 0.69 V on polycrystalline Cu ^[86] and 0.24 V on Au electrodes. ^[28] According to our simulations, the CO production shows a reasonable overpotential of 0.56 V for configuration SA2 on Al‐C₃N₄. The overpotentials for other hydrocarbon productions (CH₃OH and CH₄) show a little increased η values around 1 V. Hence, compared with literature values on the CO₂ reduction on transition metals[ ¹⁶ , ²⁵ , ⁴⁰ , ⁴² , ⁴³ , ⁹⁴ , ⁹⁵ , ⁹⁶ ] such as M‐C₃N₄ (M=Cu, Ag, Au, Ni, Co, Fe, and Pt), the present Al‐C₃N₄ and P‐C₃N₄ SACs demonstrate improved activity and reactivity for electrochemical reduction of CO₂ to simple C1 products.

Conclusions

We explored the geometrical surfaces, stabilities, catalytic activities, and selectivity of 2D M‐C₃N₄ (M=Al and P) catalysts as the embedded single‐atom electrocatalytic materials for CO₂ reduction reactions by performing periodic DFT computations. The computational model of the hydrogen electrode was employed to explore the preferable electrochemical reduction mechanisms. Computations showed that the proposed SACs are thermodynamically stable, and CO₂ can be adsorbed by not only the single atoms but also surface FLPs (Al/N or P/N). Extensive computational simulations demonstrated that CO₂ is more likely to be reduced to formic acid with low overpotentials 0.25 and 0.08 V on Al‐SAC and P‐SAC, respectively. Besides, these 2D surfaces experience the potential determining steps to produce HCOOH as 0.50 and 0.33 eV on Al‐C₃N₄ and P‐C₃N₄, respectively. While other C1 products (CO, H₂CO, CH₃OH, and CH₄) can also be yielded by CO₂RR on these catalytic surfaces, the production of methane and methanol on P‐C₃N₄ is preferred due to low limiting potentials of about 0.9 eV.

However, we note that Al‐SAC may have a low overpotential for the HER, which would suggest it lacks selective activity for the CO₂RR. But the strong adsorption energy of CO₂ on the single metal site leads to the competition between CO₂ and H for the single metal site. Still, it is essential to consider that the low overpotential for HER might still pose a challenge to achieving high selectivity for CO₂RR. We expect that our present computational study potentially benefits the design of low‐cost and environment‐friendly catalysts by using Earth abundant elements instead of precious transition metals.

Conflict of Interests

The authors declare no conflict of interest.

Supporting information

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer reviewed and may be re‐organized for online delivery, but are not copy‐edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors.

Supporting Information

Data Availability Statement

The data that support the findings of this study are available in the supplementary material of this article.