Atomically Dispersed Alkaline‐Earth Metals as Active Centers for CO2 Electroreduction to Exclusively Produce Formate

Electrocatalytic CO2 reduction reaction (CO2RR) to produce formate (HCOOH) attracts special interest in the upgrade of waste CO2. For the selective CO2 conversion into HCOOH, the preferable binding of *OCHO compared with *COOH is a prerequisite, which presents a great challenge to the rational design of the catalytic active center. Recently, alkaline‐earth (AE) metals as active centers have been reported for electrocatalysis. Herein, the feasibility of AE metals as active centers in heterogeneous catalysis for electrocatalytic CO2RR toward HCOOH based on a series of AE metal single‐atom catalysts (SACs) is theoretically studied. High‐throughput first‐principles calculations reveal that, for all the studied systems, the AE metal active centers preferably adsorb *OCHO, enabling exclusive HCOOH production. Especially, Mg SACs embedded in graphene and Ca SAC anchored in g‐C2N can efficiently convert CO2 into HCOOH under near‐zero potential, and both systems exhibit high stability. Mechanistic investigation indicates that the AE metal active centers are highly ionic, which can strongly bind *OCHO mainly through the electrostatic attraction interaction. This study lays a theoretical foundation for the rational design of AE metal SACs for efficient CO2 electroreduction with exclusive HCOOH selectivity, and further emphasizes the potential of AE metals as active centers in heterogeneous catalysis.

community. [26,27] Different from the localized d orbitals of transition metals, the delocalized s/p orbitals play an indispensable role in the catalysis based on the p-block metals. And interestingly, it is shown that In-, Sn-, and Bi-based catalysts display excellent performance for CO 2 RR to produce HCOOH. [7,[14][15][16]18] On the contrary, alkaline-earth (AE) metals (Mg, Ca, Sr, Ba, etc.) with low electronegativity are usually bivalent in their compounds and exist in ionic form, due to that they are easy to lose their two s valence electrons. Consequently, AE metals are often considered to be of low activity in heterogeneous catalysis due to the lack of a combination of empty and filled host orbitals for charge exchange. [28,29] However, surprisingly, very recently several works uncover the potential of AE metals as active centers for heterogeneous catalysis. [20,[30][31][32][33] For example, Liu et al. demonstrated that the Mg single-atom catalyst (SAC) embedded in the N-modified graphene exhibits a strikingly high activity for 4e À oxygen reduction reaction with a half-wave potential of 910 mV. [32] Combining experiments and theoretical calculations, we showed that the Mg single atom in 2D layered Mg 3 (hexaiminotriphenylene) 2 can efficiently reduce O 2 into H 2 O 2 with selectivity over 90%. [30] As described earlier, the catalyst surface that binds *OCHO more strongly than *COOH would selectively hydrogenate CO 2 to HCOOH. Moreover, for both *OCHO and *COOH, due to the electronegativity difference, generally C atom will carry a positive charge, while the O atoms carry a negative charge. Considering that *OCHO binds with the active site through O atoms, while *COOH tends to bond to the active site through C atoms, it is expected that the AE metal cations as active sites are capable of preferentially binding *OCHO ( Figure 1a) and selectively producing HCOOH, which remains to be explored.
In this work, taking SACs with AE metals (Mg and Ca) embedded into N-modified graphene and graphited carbon nitride monolayers as representatives (AE-SACs), we theoretically explored the potential of AE metal as active centers for CO 2 RR to selectively produce HCOOH. Intriguingly, the AE metal active centers can strongly adsorb *OCHO but hardly adsorb *COOH, thus leading to the exclusive production of HCOOH. Furthermore, the binding strength of *OCHO can be properly regulated to achieve excellent CO 2 RR activity by tuning the metal atoms at the active centers and their coordination environments. Especially, the Ca SAC anchored in g-C 2 N exhibits ultrahigh CO 2 RR activity with near-zero limiting potential (U L ) of -0.01 V, and the Mg SAC coordinated with two C and two N atoms can even accomplish the CO 2 RR to HCOOH without any applied potential. Moreover, the underlying mechanism for the AE metal active centers to exclusively convert CO 2 to HCCOH was explored.

Computational Details
All the spin-polarized density functional theory (DFT) calculations were performed by the Vienna ab-initio simulation package (VASP) [34] with the generalized gradient approximation (GGA) parameterized by Perdew-Burke-Ernzerhof (PBE) exchangecorrelation functional. [35] The projector-augmented wave (PAW) method was applied to describe the electron-ion interaction with a plane-wave cutoff energy of 450 eV. The convergence thresholds during the structural relaxation were set to 10 À5 eV for total energy and 0.02 eV Å À1 for Hellmann-Feynman force, respectively. The 5 Â 5 Â 1 supercell for graphene and 2 Â 2 Â 1 supercell for g-CN, g-C 2 N, and g-C 3 N 4 were adopted as supports to construct SACs with a 20 Å vacuum thickness along the z direction. The effect of relaxing lattice constants on the adsorption stability was tested, and it is found that thanks to the large supercells relaxing the lattice parameters have negligible influence on the adsorption energy ( Figure S1, Supporting Information). The Brillouin zone was sampled by a Γ-centered Monkhorst-Pack scheme with k-points grids of 3 Â 3 Â 1 and 6 Â 6 Â 1 for structural optimization and density of states (DOS) calculation, respectively. Grimme's DFT-D3 method was incorporated to deal with the van der Waals correction. [36] The thermal stability was evaluated through the ab-initio molecular dynamics (AIMD) simulations within the NVT ensemble. [37] Bader charge analysis was executed to quantify the charge transfer. [38] To explore the accessibility of experimental synthesis of the graphene-based AE-SACs, the formation energy (E f ) was calculated by the following formula [39,40] where E total and E G are the total energies of the AE-SAC system and 5 Â 5 graphene supercell, respectively. μ i and n i represent the chemical potential and number of the i species, respectively. n i is the number of the removed or added i species, where the positive and negative values are referred to the removed and added species, respectively. The chemical potentials of the involved species, μ C , μ N , and μ AE , were calculated using a pristine graphene unit cell, free N 2 molecule, and the AE metal bulk, respectively. Under this definition, the lower E f means that the AE-SAC is easier to be synthesized. [41] To estimate the binding strength of a single AE atom with the substrate, the binding energy (E b ) was expressed as where E total and E substrate represent the total energy of the AE-SAC system and bare substrate, and E AE is the total energy of isolated AE metal. The more negative E b implies higher stability of the AE-SAC. [42,43] According to the computational hydrogen electrode (CHE) model proposed by Nørskov et al, [44] the Gibbs free energy change (ΔG) for each elementary step was calculated by in which ΔE, ΔE ZPE , and ΔS are the change of total energy, zeropoint energy, and entropy, respectively. T is the temperature of 298.15 K. The zero-point energy and entropy can be calculated from vibrational frequencies for intermediates or acquired from the NIST database [45] for free molecules, and corresponding values are listed in Table S1, Supporting Information. As wellknown, the CHE model could well describe the electrochemical reaction mechanism and predict the reaction activity and selectivity trend, thus has been widely used to study various electrochemical reactions. [46,47] Especially, for CO 2 RR, the experimental observations can also be well reproduced by the theoretical calculations based on the CHE model. [15,48,49] The limiting potential was defined as U L ¼ -ΔG max /e, where ΔG max is the free energy change of the potential-determining step (PDS). [50] The previous study has demonstrated that under the frame of the CHE model, the pH changes have no influence on the reaction overpotential, so were also not considered in this work. [23] The solvation correction was carried out based on previous studies, [51][52][53][54] that is, the intermediates with hydroxyls bonded to active sites directly and indirectly are stabilized by %0.5 and 0.25 eV, respectively, and those containing no hydroxyl are stabilized by 0.1 eV. In fact, we also calculated the solvation correction of considered systems with the implicit solvation model as implemented by VASPsol. [55] The obtained solvation correction for *OCHO and *COOH are -0.10 and -0.22 eV, respectively, agreeing reasonably with the empirical values, thus the empirical solvation correction is suitable for our systems. Considering that the electrocatalytic CO 2 RR always takes place in solution, according to a previous study, [52] the liquid HCOOH was selected as the target product with a liquid phase free energy correction of -0.22 eV.

Results and Discussion
Due to the unique advantages of facile fabrication, excellent stability, easily tunable geometric and electronic structure, and so on, graphene-based SAC has been widely used for various electrocatalytic reactions, which is also regarded as an ideal model system to understand the fundamental physical chemistry of SACs, especially appropriate for theoretical studies. [56,57] Very recently, several SACs with AE metals as active centers have been successfully synthesized and used for heterogeneous catalytic reactions. [20,[30][31][32][33] Considering that the electronic structures and catalytic performances of SAC could be effectively turned by regulating the coordination environment of its active center, [58][59][60] herein, we built a series of SACs by embedding single Mg/Ca atom into graphene with double vacancy decorated by various numbers of N atoms. In addition, three typical graphited carbon nitride monolayers, i.e., g-CN, g-C 2 N, and g-C 3 N 4 , which have been widely employed as substrates for SACs, were also taken into account. [24,42,43,52,61,62] Figure 1b displays the schematic configurations of the AE-SACs with Mg/Ca metal anchored on these porous N-coordinated carbon nanosheets. By regulating the numbers of coordinated N atoms, 6 graphenebased substrates were constructed to support the Mg or Ca metal. With the addition of three carbon nitride monolayers as substrates, a total of 18 SAC candidates were obtained.
The key parameters of studied AE-SACs are listed in Table S2, Supporting Information. To evaluate the synthetic feasibility of the graphene-based AE-SACs, their formation energies have been investigated and presented in Figure 2a. It can be seen that the calculated formation energies of all graphene-based AE-SACs (except MNC 3 systems) are lower than or comparable to that (4.02 eV) of MgC 2 N 2 -III SAC. Given that MgC 2 N 2 -III SAC has been synthesized in the experiment, these systems can also be fabricated under suitable experimental conditions. [32] The binding energies have been further studied for graphene-based AE-SACs. From Figure 2b, the large negative binding energies (<-3 eV) mean the Mg/Ca atom can be strongly anchored with the graphene substrates, indicating that there are strong chemical bonds formed between the embedded AE metal and its neighboring coordinated atoms in the substrate, and the aggregation of AE metals can be effectively inhibited. Furthermore, for the 2D holey carbon materials, the binding energies were also used to evaluate the synthetic feasibility, and the large negative binding energies ensure their experimental synthesis and atomic dispersion, similar to the cases of TM-based SACs. [63,64] For the electronic structure, Bader charge analysis presented in Figure 2c shows that significant electrons have transferred from the anchored metal atoms to the substrates (>%1.4 e), indicating that these AE metal active sites are highly positively charged. It is worth mentioning that the valence states of the AE metals are generally considered as "þ2" in their compounds. [65] However, here the "þ2" means the "formal oxidation states" [66] of AE metals, but not the real charge amount. In fact, the theoretical value of the charge amount of the atom in a compound heavily depends on the charge partition method. As a relevant case, Bader charge analysis shows that the Cs atoms can transfer 0.83, 1.32, 1.93, 2.34, 2.89, and 3.18 electrons to F atoms in CsF, CsF 2 , CsF 3 , CsF 4 , CsF 5 , CsF 6 , respectively, which have the formal oxidation states of "þ1", "þ2", "þ3", "þ4", "þ5", "þ6". [67] In view of this, it can be reasonably inferred that the Mg and Ca atoms can be considered as formal "þ2" valence even though transferring %1.6 and %1.5 electrons to the substrates, respectively. Taking MgC 2 N 2 -III and Ca/g-C 2 N as examples, electron localization function (ELF) in Figure 2d further demonstrates the electron-deficient characteristics of the AE metal active centers, which can favor the adsorption of *OCHO as shown below. Then we studied the DOS, as displayed in Figure 2e and S2, Supporting Information. The total DOS of these AE-SACs display metallic or semimetallic characteristics, which is beneficial to the charge transfer during CO 2 RR, and the projected DOS of the anchored AE metals are in line with the charge transfer analysis, from which we can see the electron states are almost empty below the Fermi level. As discussed earlier, the first protonation step of CO 2 plays a critical role in the entire CO 2 RR process, because the generated intermediates, *COOH and *OCHO, will bring about entirely different CO 2 RR pathways as well as products. [8][9][10][11][12] All the initial possible configurations of *COOH and *OCHO on the constructed SACs have been considered, and the most stable ones are presented in Figure 3 and S3, Supporting Information, for which the key structural parameters are listed in Table S3-S5, Supporting Information. It is noted that for *OCHO, the formate radical is bound to the active site through two O─AE metal bonds, while for *COOH, the carboxyl radical bonds to the AE active center through either one C─AE metal bond or combining the C─AE metal and O─AE metal bonds. As expected, *OCHO binds more strongly than *COOH as suggested by the much more negative ΔG *OCHO than ΔG *COOH in    Figure 3c. Thus, the formation of *OCHO is much more favorable than *COOH when CO 2 is hydrogenated on all the studied AE-SACs, which suggests that HCOOH would be the preferred product for CO 2 RR on the AE-SAC systems. Generally, the CO 2 RR proceeds in an aqueous environment and involves proton transfer, and the hydrogen evolution reaction (HER) is the main competing reaction. [68,69] To ensure the active center is exclusive to CO 2 RR, the competitive adsorption of a proton should be excluded, which requires stronger adsorption of *OCHO/*COOH than *H. As seen in Figure 3c, for all the systems the adsorption-free energies for *H are highly positive, while, on the contrary, *OCHO are highly negative. Consequently, the AE metal active sites will be preferentially occupied by *OCHO rather than *H, thereby suppressing HER.
Since the hydrogenation of CO 2 prefers to form *OCHO on all AE-SAC systems, next, we focus on exploring the reduction process of *OCHO by evaluating free energies. All possible hydrogenation positions of intermediates during the CO 2 RR have been considered, and only the most stable structure in each elementary step is chosen to further study the subsequent reaction for economically and effectively obtaining the optimal pathway. The U L values of the AE-SAC systems were calculated and shown in Figure 4a and Table S6, Supporting Information. Interestingly, a majority of studied AE-SACs exhibit high CO 2 RR activity with less negative U L than -0.5 V, going against the conventional wisdom that the AE metals are catalytically inactive. [28,29] Especially, Ca/g-C 2 N possesses an excellent CO 2 RR activity with a U L near-zero (-0.01 V), what is more, it even does not require any applied potential to proceed with the CO 2 RR on MgN 2 C 2 -III, i.e., all the electrochemical steps are exothermic. Note that the above U L values are much lower than that of typical TM-based SACs. [62,64,[70][71][72][73][74][75] For instance, the U L for Ni/WTe 2 , Cu/g-CN, Mn/Ti 2 CN 2 , FeN 2 C 2 /G, Fe/GDY, and Ni/g-C 3 N 4 are À0.11, À0.28, À0.32, À0.38, À0.39, and À0.74 eV, respectively. In addition, the AE-SACs would be also better than the recently fabricated p-block metal-based SACs, including InN 4 /G, [15] SbN 4 /G, [16] and SnN 2 C 2 /G, [76] which have the theoretical U L of -0.71, -0.21, and -1.02 V, respectively. Considering that MgN 2 C 2 -III has been synthesized and applied for oxygen reduction reaction, [32] it is expected that our theoretical prediction can be verified in the future.
According to the Sabatier principle, either too strong or too weak adsorption of the reaction intermediates is not conducive to the whole catalytic process. [57] Moreover, the adsorption-free energy of the key intermediate has been widely used as the descriptor in theoretical electrocatalysis to predict the activity trend and guide the catalyst design for various reactions. [19,62] Hence, we examined the relationship between the adsorptionfree energy of *OCHO and the U L for the AE-SAC systems. As shown in Figure 4b, there are good linear relationships between the ΔG *OCHO and U L for the considered AE-SAC  systems. Interestingly, we can see that the weaker adsorption of *OCHO could lead to a higher CO 2 RR catalytic activity to produce HCOOH. In addition, Ca-based SACs (slope of 1.12) exhibit different trends compared with Mg-based SACs (slope of 0.44), which may be related to their different adsorption strength of *OCHO. As seen in Figure 3c, for the same coordination environment, Ca-based SACs always possess more negative ΔG *OCHO than Mg-based SACs on graphene-based substrates, but the aforementioned situation is completely reversed on the 2D holey carbon materials. And the fluctuation of ΔG *OCHO for Ca-based SACs (-1.26 % -0.47 eV) is much smaller than that for Mg-based SACs (-1.38 % -0.10 eV).
To get more insight into the reaction mechanism for CO 2 RR to HCOOH, the detailed reaction pathways on MgN 2 C 2 -III and Ca/g-C 2 N are plotted in Figure 4c,d, respectively, and those for other systems are depicted in Figure S4, Supporting Information. The first hydrogenation steps of CO 2 prefer to form *OCHO with the free energy downhill of 0.10 and 0.47 eV on MgN 2 C 2 -III and Ca/g-C 2 N, respectively, while for *COOH the free energies rise by 1.39 and 0.91 eV. For the second hydrogenation step, it will consume more energy for the proton-electron pair to attack the middle C of *OCHO than to attack side O of *OCHO, thus the *HCOOH are preferentially formed on both MgN 2 C 2 -III and Ca/g-C 2 N with free energy downhill of 0.17 eV and uphill 0.01 eV, respectively. Then, the ΔG for further hydrogenation on the C or O sites of *HCOOH are much more positive than the ΔG for HCOOH desorption from the aforementioned two systems, indicating the desorption of *HCOOH is easier than its further hydrogenation. Hence, it is predicted that HCOOH can be exclusively produced by CO 2 RR on MgN 2 C 2 -III  and Ca/g-C 2 N with negligible potential, and their PDSs are * þ CO 2 þ H þ þ e À ! *OCHO and *OCHO þ H þ þ e À ! *HCOOH, respectively. Interestingly, the PDSs for CO 2 RR to HCOOH on the other AE-SAC systems are all the second hydrogenation step, i.e., *OCHO þ H þ þ e À ! *HCOOH.
To get fundamental insight into the physical mechanism of much stronger adsorption of *OCHO than *COOH, which enables exclusive production of HCOOH for CO 2 RR, the electronic properties of MgN 2 C 2 -III and Ca/g-C 2 N with *OCHO/*COOH were analyzed in terms of PDOS, crystal orbital Hamilton population (COHP), and ELF. It is observed from the PDOS in Figure 5a,b that compared with the TM active center, [77][78][79] the occupied electronic states of Mg and Ca are very few for both *OCHO and *COOH, indicating that the bonding interactions between Ca/Mg and *OCHO/*COOH are mainly dominated with the ionic bonds, completely different from the predominant covalent bond interaction between TM and *OCHO/ *COOH. [77][78][79] Bader charge analysis was performed to quantify the charged state of *OCHO and *COOH. As seen in Table S7, Supporting Information, except the O atoms are negatively charged, the Mg/Ca and C atoms are positively charged, which plays a major role in the strong adsorption of *OCHO compared with *COOH. Moreover, COHP analysis (Figure 5a,b) shows that there are also contributions from the covalent bond. For both MgN 2 C 2 -III and Ca/g-C 2 N, the integrated COHP (ICOHP) values for *OCHO (-4.91 and -4.16 for MgN 2 C 2 -III and Ca/g-C 2 N, respectively) are more negative than *COOH (-2.78 and -3.12 for MgN 2 C 2 -III and Ca/g-C 2 N, respectively), indicative of the stronger covalent interaction between *OCHO and the AE active centers than *COOH, which also makes a contribution to the preferred adsorption of *OCHO compared with *COOH. The above bonding picture can be further qualitatively confirmed by the ELF presented in Figure 5c,d, from which we can see that electron is significantly deficient on Ca or Mg, while there is some electron accumulation between *OCHO/*COOH and Ca/Mg. Earlier, the negative binding energies have certified the thermodynamic stability of MgN 2 C 2 -III and Ca/g-C 2 N, then the AIMD simulations were performed to check their thermal stability. As illustrated in Figure 6, the temperature and energy curves oscillate around the equilibrium states, and the geometric structures of MgN 2 C 2 -III and Ca/g-C 2 N retain well after a 10 ps AIMD simulation under 600 K, suggesting they are stable even at high temperature.
In addition, we also calculated the adsorption free energy of H 2 O (ΔG *H 2 O ) and dissolution potential (U diss ) of the AE-SACs to evaluate their stabilities under moister conditions. As seen in Table S8, Supporting Information, the values of ΔG *H 2 O for most systems are higher than -0.6 eV, indicating the weak interaction between the H 2 O molecule and the AE-SACs ( Figure S5, Supporting Information). Although Mg/g-CN and Mg/g-C 2 N interact with H 2 O strongly, the CO 2 RR activities of both systems are very poor. The weak adsorptions of H 2 O on the AE-SAC systems signify that the H 2 O molecule could facilely desorb from the catalysts. Then the U diss of the studied AE-SAC systems is calculated according to the formula described in previous work. [50] It can be seen from Table S8, Supporting Information, MgN 2 C 2 -III has a U diss of -0.90 eV. Considering that this system has been experimentally synthesized and used in the oxygen reduction reaction, [32] most of the AE-SACs are also stable enough under the electrochemical condition, due to their less negative or comparable U diss values. Especially for Ca/g-C 2 N, the U diss of -0.47 V indicates that it even has higher electrochemical stability than the experimentally synthesized MgN 2 C 2 -III system. [32] In total, the weak H 2 O adsorption and acceptable electrochemical stability could further reasonably indicate the good stabilities of the AE-SAC systems under moister conditions.
Finally, we also explore the possible preparation methods of the AE-SAC systems. Drawing on a similar synthetic technique of MgN 2 C 2 -III and CaN 4 systems, [31,32] the graphene-based AE-SAC systems can be prepared via the pyrolysis strategy by direct pyrolysis the proper Mg/Ca-contained metal-organic-framework or complexes and then acid pickling. The wet-chemistry strategy, which has been widely used for the preparations of transition metal-based SACs, [80,81] could also be adapted to prepare the graphited carbon nitride monolayer (g-CN, g-C 2 N, g-C 3 N 4 ) supported AE-SACs. Briefly, the g-CN, g-C 2 N, g-C 3 N 4 supports are first mixed with suitable Mg/Ca-based salts, such as CaCl 2 or MgCl 2 . Next, collect, wash, and then heat-treat the aforementioned mixtures at a certain heating rate to acquire corresponding target products.

Conclusions
To sum up, based on first-principles calculations, a series of AE-SACs (AE ¼ Mg and Ca) were constructed to explore the feasibility of AE metals as active centers for CO 2 RR to produce formate (HCOOH). Intriguingly, the AE metal active centers in all studied SAC systems highly prefer to adsorb *OCHO compared with *COOH, leading to the exclusive production of HCOOH. By thoroughly studying the reaction mechanism, the MgN 2 C 2 -III and Ca/g-C 2 N SACs were identified as the optimal candidates, which can accomplish the CO 2 RR to produce HCOOH under theoretical near-zero limiting potential, better than the overwhelming majority of TM-based SACs. In addition, both systems also exhibit good stability and excellent selectivity against the competing HER. Electronic structure analysis indicates that the AE metals are highly ionic when forming SACs, and the positively charged Mg/Ca active centers prefer to bind the negatively charged O in *OCHO due to the electrostatic interaction, which contributes to the observed excellent activity and selectivity for producing HCOOH. This work provides a useful guidance for the rational design of the atomic active site for selectively producing valuable HCOOH via electroreduction of CO 2 , and highlights the great promise of the alkaline-earth metal as active centers in heterogeneous catalysis. It is noted that MgN 2 C 2 -III SAC has been synthesized and used for the 4e À oxygen reduction reaction, [32] and thus it is expected that the design principle proposed herein can be experimentally verified in future.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.