The ever-rising global energy consumption and its negative impact on the environment are major challenges in front of humanbeing in the 21st century1,2,3,4,5,6,7. Public health is severely threatened by the greenhouse effect due to the excessive fossil fuel usage. It is highly desirable to develop technologies that can convert the greenhouse gas (i.e., CO2) into sustainable and clean energy sources. Among various possibilities, catalytic reduction of CO2 into hydrocarbon fuels, which features environmental friendliness, high efficiency, and low cost, has been recognized as one of the most promising approaches. Single atom catalysts (SACs), first proposed for CO oxidation in 20118, offer us excellent candidates to activate and convert CO2 efficiently when trasition metal (TM) atoms are used9,10,11,12,13, because the coexistence of empty and occupied TM d-orbitals can accept lone-pair electrons, and then back-donate these electrons to the antibonding orbitals to weaken the C=O bonds. Notably, two-dimensional (2D) materials serve as promising substrates for atomically dispersed transition metal atoms for CO2 reduction due to their large surface-to-volume ratios, short carrier diffusion distances, unique electronic properties, and abundant active sites. The catalytic performance of these 2D SACs stems from versatile advantageous characteristics. For example, catalysts with singly dispersed Ni, Co, Cu, Pt, and Pd atoms on graphene14, Pt or Pd atom anchored on C3N415, V-β12 boron monolayer16 and Co decorated metal–organic frameworks17,18 all have shown great potential for CO2 reduction.

Despite latest advances, there is still plenty of room to improve the catalytic activity and selectivity of CO2 reduction in order to satisfy large-scale industrial applications. Recent studies have shown that electric polarization plays an important role in determining catalytic activities and selectivities, and can greatly improve the catalytic performance compared with the unpolarized counterparts19,20,21. For instance, Janus TM dichalcogenides possess better catalytic abilities for water splitting and N2 fixation19,20,21, since the catalysis-related properties, including surface stoichiometry, electronic structure22,23,24,25,26, adsorption strengths, and reaction activation energies, can be tuned by the polarization27; Moreover, by utilizing the polarization from the substrate, the overpotential of oxygen evolution reaction on TiO2 surface can be strongly reduced28.

Inspired by these recent achivements, we expect that ferroelectrics, viz., materials with switchable ferroelectric polarization, hold promise as efficient catalysts since their unique reversible polarization provides an additional mechanism to adjust empty and occupied d-orbitals of adsorbed metal atoms. The activation of reactants can thus be controlled, while the catalytic properties of the supported metal, such as activity29,30,31,32 and selectivity, can be tuned33. These expectations have been partially confirmed in ABO3 perovskite ferroelectrics. For instance, reversing the polarization of ferroelectric PbTiO3 substrate makes CrO2 monolayer overcome the limiting factor stemming from the Sabatier principle, and thus display excellent catalytic behaviours for both NOx decomposition and CO oxidation34. Reorienting the polarization direction of the ferroelectric PbTiO3 substrate can dramatically change the chemisorption energies of CO, O, C, and N on the PbTiO3-supported Pt films, and alter the reaction paths of the dissociations29. The oxygen evolution reaction activity of TiO2 film on a ferroelectric SrTiO3 substrate can be strongly enhanced relative to unsupported TiO2 due to the presence of dynamic dipoles in response to the charge on the adsorbed species, while the reaction path can be modulated via ferroelectric switching28. However, these predictions have not been experimentally realized so far, mainly due to the instability and depolarization in traditional ABO3 perovskite ferroelectrics when the thickness is smaller than a critical value. Encouragingly, the emergence of 2D ferroelectric materials, exemplified by In2Se335, CuInP2S636, and SnTe37, provides an excellent opportunity to verify controllable catalysis in ferroelectrics. The realization of ferroelectricity at room temperature makes it feasible for 2D ferroelectrics to be used for controllable catalytic CO2 reduction.

In this work, based on the experimentally available 2D ferroelectrics, we theoretically investigate the potential of using transition metal decorated α-In2Se3 monolayers as the ferroelectric SACs toward electrochemical CO2 reduction. Our efforts identify six catalysts with singly dispersed TM atoms anchored on the 2D ferroelectric substrate, namely α-In2Se3 monolayer, with polarization downwards or upwards (denoted as TM@P ↓ -In2Se3 or TM@P ↑ -In2Se3). The switchable polarization can not only alter the reaction barrier and paths of CO2 reduction, but also lead to different final products. It can even reactivate the stuck CO2 reduction. These performance improvements stem from the synergistic effects of adjusted empty and occupied d-orbitals (d orbital center) of adsorbed metal atom, polarization dependent electron transfer, and CO2 adsorption energies under ferroelectric switching. These ferroelectric SACs, catalytic mechanism, and the polarization dependent catalytic CO2 reduction open an avenue for controllable CO2 reduction reaction (CO2RR), and introduce a feasible approach to significantly improve the efficiency of such reactions.


Screening ferroelectric catalysts

The ferroelectric α-In2Se3 monolayer was chosen as the substrate of potential catalysts due to the following reasons: (1) it is the ground-state phase among possible phases (see Supplementary Fig. 1), even in the presence of adatoms and in electrochemical environment (see Supplementary Fig. 2 and Supplementary Tables 12), which is consistent with the recent theoretical finding by Ding et al.38 Note that two α-In2Se3 (α and α’) phases have been predicted to share similar atomic structures and are almost energetically degenerate, and they have the same CO2RR performance (Supplementary Fig. 3). We thus mainly focus on α- In2Se3 in this study. (2) it is stable at room temperature and has been experimentally synthesized35,39, and its ferroelectricity with switchable polarization under room temperature has been explicitly demonstrated35,40. With the polarization locking from the asymmetric arrangement of its quintuple layers (see Fig. 1a, b), their orientations can be reversed by the shift of the middle Se layer38. However, as a ferroelectric semiconductor with a sizable band gap (1.46 eV)38, α-In2Se3 monolayer does not have enough electrons to be injected into the antibonding 2πu orbitals of CO2 so that the strong sp-hybridization symmetry of the carbon atom can not be disrupted41. Thus, the material itself is not suitable as a catalyst for CO2 reduction42, which is corroborated by our theoretical study: upon adsorption on In2Se3, the inherent linear O=C=O structure of CO2 molecule is well maintained (see Supplementary Fig. 4 and Supplementary Table 3).

Fig. 1: Geometries and electronic structures of the ferroelectric catalysts.
figure 1

Top and side views of the optimized α-In2Se3 monolayer with a downward (P ↓ ) and b upward (P ↑ ) polarization. Red circles denote selected adsorption sites: I, top of the In atom; C, center of the six-membered ring; B, top of the In-Se bond; S, top of the Se atom. The black rhombus represent the unitcell of In2Se3. c The d band center of TM@In2Se3 (TM = Ni, Pd, Rh, Zr, Nb and Re). Top and side views of the optimized configuration of d Pd@P ↓ -2×2 In2Se3 and e Pd@P ↑ -2×2 In2Se3, the supercell is indicated by the red rhombus. f The partial density of states of Pd@P ↓ -In2Se3 and Pd@P ↑ -In2Se3. The insets show the 3D differential charge density plots obtained with the model shown in Fig. 1d and e. The isosurfaces are 0.005 e/Å3. Charge accumulation and depletion are marked by the yellow and blue regions, respectively.

To activate electrochemical CO2 reduction, we introduced transition metal atoms to decorate α-In2Se3 monolayer, which provide extra electrons to break the strong sp hybridization, thus activating the CO2 molecules29,42,43,44. Compared with conventional metal catalysts, SACs normally exhibit higher catalytic performances because of the high ratio of low-coordinated configurations8,45,46. However, the transition metals have to be screened since only the SACs with well balanced empty/occupied d orbitals can exhibit optimal catalytic performance. To find suitable SACs, 29 transition metal atoms were chosen and chemically adsorbed (see Supplementary Fig. 5) on both surfaces of the In2Se3 monolayer, and the energetically most favorable configurations were identified. The possible metal substitutions into the In2Se3 monolayer were excluded due to the large formation energy of In vacancy, high diffusion barriers of In atoms (see Supplementary Fig. 6), and low CO2RR activities (see Supplementary Fig. 7). Then we evaluated all the composites based on two key criteria. First, to ensure the stability of the active site, the single transition metal atom should be steadily adsorbed on the In2Se3 monolayer surface without breaking the underlying structure, and the favourable adsorption site should not change significantly after the polarization switching; second, the single transition metal atom should be able to activate CO2, in other words, the linear structure of O=C=O should be broken after adsorption, at least in one polarization state.

After a comprehensive investigation of the energetically most favorable adsorption configurations (Fig. 1a, b) and their stabilities, we screened out six TM atoms (TM = Ni, Nb, Pd, Re, Rh and Zr) as promising catalysts toward CO2 reduction for further examinations. Other 23 candidates are either unstable or cannot effectively activate CO2 molecules (see details in the Supplementary Information).

Stability of SACs and polarization dependent d band center

Although the screened TM@In2Se3 are stable (without structural breaking) and able to activate CO2 molecules, transition metal atoms need to uniformly disperse on the substrate to prevent any aggregation of the deposited atoms and maximize the catalytic efficiency47.

To achieve this goal, strong binding strength, high diffusion energy barrier, and negative clustering energy are essential15,48,49,50,51,52,53,54. At the energetically most favorable site, all the six TM atoms have rather strong binding energies (Table 1) that are comparable to previously predicated SACs15,49,55,56, indicating sufficiently strong interactions between the metal atoms and the substrate, as evidenced by the strong TM–Se bonds (Fig. 1d, e and Supplementary Fig. 8). To provide the reference for the binding strengths, we compared the binding energies of TM atoms in ferroelectric SACs with the ones in the corresponding molecular precursors, it is found that the adsorption of TM (e.g. Pd, Rh and Ni) atoms on In2Se3 is energetically preferred, indicating the feasibility of the SACs to be synthesized based on the corresponding molecular precursors (see Supplementary Table 4). The chemical bonding interaction is illustrated by the substantial partial charge accumulation between the TM atoms and the surrounding Se atoms (see Table 1, and Fig. 1f and Supplementary Fig. 8) as well as the strong hybridization of Se p and TM d orbitals (shown in Fig. 1f and Supplementary Fig. 9) that is similar to the case of TM atoms adsorbed on 2D carbo-nitride15,49,55. Moreover, the high diffusion barriers of TM atoms on the surfaces can effectively prevent the formation of metal clusters (see Supplementary Fig. 10), though specific values depend on the ferroelectric polarization.

Table 1 Parameters for the pure TM@In2Se3 and CO2 adsorbed TM@In2Se3: adsorption site (Sad), binding energies of TM atom (Eb-TM in eV/atom) and CO2 molecule (Eb-CO2 in eV/molecule), charge lost from the adsorbed TM atoms (QTM in e/atom), average TM-Se bond length (lTM-Se in Å), migration barrier (Eba-TM in eV), charge gained by the adsorbed CO2 molecule (QCO2 in e/molecule), and bond angle (OCO in °).

To further demonstrate the structural stability of the SACs chosen in this work, we also conducted clustering energy calculations and ab initio molecular dynamics (AIMD) simulations with a Nose–Hoover thermostat at 300 K on Pd@In2Se3 as a representative example. The neagive clustering energies (−0.3 eV and −0.04 eV for Pd@P ↓ -In2Se3 and Pd@P ↑ -In2Se3, respectively; see details in SI) indicate that it is not energetically favorable to form clusters on the catalyst surface. In the AIMD simulations, the out-of-plane polarization of ferroelectric In2Se3 monolayer can be well retained, and the metal atom stays anchored at the energentically favorable site even at room-temperature (300 K) for at least 15 ps (see Supplementary Fig. 11a). Besides, AIMD simulations at 300 K are also performed to investigate two dispersedly adsorbed Pd atoms on P ↓ -In2Se3 (2 Pd@P ↓ -In2Se3). The two Pd atoms can maintain dispersedly adsorbed features without metal clustering and structural phase transition for 15 ps (see Supplementary Fig. 11b). The distance between the two Pd atoms stays almost unchanged (as shown in the inset of Supplementary Fig. 11). The possible metal (e.g. Pd and Nb) agglomerations are excluded by the further kinetic Monte Carlo (kMC) simulations, the clusters will not form at the surface for 100 seconds at the room-temperature (see Supplementary Figs. 1214). Overall, the strong adsorption energies, high diffusion barriers, negative clustering energies, molecular dynamics simulation, and kMC results all indicate strong stability of the 2D ferroelectric SACs.

Good stabilities of the ferroelectric substrate and the catalysts themselves are key prerequisites to achieve the goal of controllable catalysis. The stability of ferroelectric α-In2Se3 monolayer has been unambiguously demonstrated by our simulations (Supplementary Fig. 1) and also by the recent theoretical and experimental studies35,36,39,40, but stabilities under harsh evironments for CO×RR still need to be addressed. To this end, based on the dissolution potential57,58,59, we have evaluated and confirmed the electrochemical stability of our proposed system under acidic conditions (see Supplementary Table 2).

Different from traditional SACs like Pd@C3N4, the switchable polarization in ferroelectric catalysts proposed herein provides an extra degree of freedom to modulate and control catalytic reactivities. The average TM-Se bond length, binding energy, and electron transfer are highly polarization dependent (Table 1). Compared with the corresponding catalysts with upwards polarization, TM@P ↓ -In2Se3 generally have shorter average TM-Se bond length, larger absolute value of binding energy, and more electron transfer, due to the large electrostatic potential difference on In2Se3 as in metal porphyrazine molecules60. The balance between empty and occupied d orbitals, caused by the polarization dependent electron transfer, significantly affect the catalytic activity of TM@In2Se3. Taking Pd@In2Se3 as an example, the charge is redistributed between the Pd and Se atoms (see Fig. 1f), and the d orbitals of Pd atom is depleted. Compared with Pd@P ↑ -In2Se3, more electron transfer occurs in Pd@P ↓ -In2Se3, shifting the Pd-d orbitals to higher energy closer to the Fermi level. As illustrated in Fig. 1c, the d band center of Pd@P ↓ -In2Se3(−1.45 eV) is higher than that of Pd@P ↑ -In2Se3 (−4.13 eV), indicating that Pd@P ↓ -In2Se3 possesses better catalytic ability46. Similar phenomena occur in the other five TM@In2Se3 catalysts (see Fig. 1c) under the same mechanism: the d band center of the TM atom is closer to the Fermi level when it is placed on P ↓ -In2Se3. Note that the position differences of the d band center under opposite polarizations for TM@In2Se3 (TM = Nb, Re) are rather pronounced, while such differences for TM@In2Se3 (TM = Ni, Pd and Rh) are only moderate. These contrasting results imply that both reaction paths and barriers are affected by the polarization (see a detailed catalytic classification below). Considering that the localized electronic states near the Fermi level lead to high N2 catalytic activity in the catalyst Co2@GDY (cobalt dimer-anchored graphdiyne)49, we expect that the modulation of the electronic structure of TM@In2Se3 caused by the polarization switching has a major influence on the subsequent CO2 hydrogenation process.

Since the catalytic activity of the TM@In2Se3 systems is polarization dependent, the effective ferroelectric switching of TM@In2Se3 is essential to achieve controllable catalytic reduction. We reverse the polarization orientation of Pd@In2Se3 (as a typical example, see Supplementary Figs. 15 and 16) via a three-step concerted mechanism on the most effective kinetic path38. We find that the energy barrier of the polarization switching after Pd decoration is 0.29 eV/UC (0.15 eV/UC) for P ↑ to P ↓ (P ↓ to P ↑ ), which is slightly larger than the value for the bare ferroelectric layer (0.13 eV/UC), but it is still accessible to experimental manipulation38. It is noted that the ferroelecitrc switching behaviors are not significantly affected by the metal adsorption. Although the local area of metal anoching site is metallized due to the hybridization between the metal and surrounding Se atoms, the semiconduting feature of the whole substrate and the associated ferroelectricity can be well preserved (see Supplementary Fig. 17). Considering that the out-of-plane ferroelectric switching of thin In2Se3 flakes has been experimentally achieved40, it should be feasible to achive similar polarization orientation reversal in TM@In2Se3 systems.

Tunable activation of CO2 by polarization switching

Effective activation of CO2 molecule is the key for the subsequent reduction. Based on the structural analysis, almost all the TM@In2Se3 catalysts could effectively activate CO2 molecule (see Fig. 2 and Supplementary Fig. 18) by forming the bidentate C-TM-O species. The only exception is Zr@P ↓ -In2Se3, where the linear O=C=O structure is well preserved (only bent by 0.8°) because of the monodentate TM-O configuration. Moreover, the binding energies of the catalysts with the activated CO2 molecule (−3.39 to −0.76 eV; see Table 1) are comparable to the reported catalysts55,61,62. The strong hybridization of O p and TM d orbitals and the significant charge transfer between TM@In2Se3 and CO2 ensure the inert molecule chemically captured (see Fig. 2b, d, and Supplementary Fig. 19). Note that the degree of CO2 activation on two opposite surfaces are obviously different as seen from the angles of OCO (see Table 1), which are generally smaller when the molecule is adsorbed on the surface with polarization downwards. Meanwhile, the activation degree of CO2 on TM@In2Se3 (TM = Nb, Re) is significantly higher than that on TM@In2Se3 (TM = Ni, Pd, Rh), as indicated by the larger binding energies (−3.39~−2.02 eV vs −1.39~−0.76 eV), more electron transfer (0.61~0.92e vs 0.26~−0.45e), and smaller angles of OCO (131.1°~136.8° vs 146.4°~151.2°).

Fig. 2: Tunable activation of CO2 by polarization switching.
figure 2

The differential charge density plots and partial density of states of CO2 adsorbed on Pd@In2Se3 (a, b for Pd@P ↓ -In2Se3, and c, d for Pd@P ↑ -In2Se3). e Simplified schematic diagrams of CO2 bonding to transition metals.

The distinct behaviours on different ferroelectric surfaces are attributed to the mechanism of CO2 catalytic activation, namely coexistence of charge depletion and accumulation between the TM atoms and CO2 molecules. The empty TM d orbital could accept electrons from the highest occupied molecular orbitals (g orbital) of the CO2 molecule, while the occupied TM d orbital could back-donate electrons to the lowest unoccupied molecular orbitals (u orbital) of the CO2 molecule (Fig. 2e). As indicated in Fig. 2b and d, the d orbital occupations of TM atoms on the ferroelectric In2Se3 layer can be effectively modulated by the switchable polarization. The synergy of electron acceptance, back-donation, and polarization dependent d orbital occupation ensures that CO2 can be efficiently activated, while the activation degree is ferroelectric controllable in TM@In2Se3 catalysts61. For example, while isolated Pd atom has fully occupied d orbital (4d10), upon adsorption on the In2Se3 substrate, the Pd atom loses partial d electrons to form the Pd-Se bond, and the resulting empty Pd d orbitals provide channels for electron acceptance and back donations to activate CO2 molecules. Consequently, compared with other TM@In2Se3 catalysts, where the TM atoms have intrinsic partially occupied d orbitals, Pd@In2Se3 catalyst has a lower degree of CO2 activation as indicated by the relatively larger OCO and smaller charge transfer (see Table 1). Furthermore, for TM@In2Se3 under polarization switching, the density of states (DOS) variations are more evident than the CO2 binding energy and electron transfer. This is due to the fact that the DOS variation is directly related with the orbital shift under the polarization flip, such as the d band center, as seen in Fig. 1c, while the charge transfer and the CO2 binding energy are mainly related to the itinerant electrons induced by the polarization.

Selectivity for CO2RR vs HER

As an important competing side reaction, hydrogen evolution reaction (HER) may significantly restrain the Faradaic efficiency of CO2RR by consuming proton–electron pairs from the electrolyte solution63,64. To check if CO2RR is more favourable, we first calculated Gibbs free energy changes (ΔG) at the first hydrogenation step of CO2RR (* +CO2 + H + +e → OCOH* or OCHO*) and HER (* +H + +e → H*). According to the Brønsted–Evans–Polanyi relation65,66, reactions with lower ΔG have smaller reaction barriers and consequently are kinetically more favoured.

As shown in Supplementary Fig. 20, for the first hydrogenation step of CO2RR, Rh@In2Se3, Pd@In2Se3, and Re@P ↑ -In2Se3 prefer to form carboxyl (OCOH*), while Ni@In2Se3, Nb@In2Se3, Zr@P ↑ -In2Se3, and Re@P ↓ -In2Se3 tend to produce formate (*OCHO). Since CO2 molecule cannot be activated on Zr@P ↓ -In2Se3, it is not considered here. The Gibbs free energy changes (Fig. 3) indicate that all the screened TM@In2Se3 catalysts have a higher selectivity toward CO2RR. These results demonstrate the feasibility of using TM@In2Se3 catalysts as cathodes for CO2RR with a high Faradaic efficiency.

Fig. 3: Selectivity for CO2RR vs HER.
figure 3

Gibbs free energy changes (ΔG) of initial protonation of CO2RR vs. HER on TM@In2Se3 (TM = Ni, Pd, Rh, Zr, Nb and Re). Data points in the purple region denote higher selectivity toward CO2RR, while those in the blue region toward HER. Black and red symbols stand for ΔG(OCOH*) vs ΔG(H*) and ΔG(OCHO*) vs ΔG(H*), respectively.

Tunable CO2RR by polarization switching

As a result of polarization dependent d band center (see Fig. 1c), CO2 molecules are activated to different degrees depending on the ferroelectric surfaces. It is thus expected that the reaction barrier, reaction path, and even the intermediate/final product of CO2 reduction will be polarization dependent and controllable via ferroelectric switching.

Considering the complexity of CO2 reaction, we comprehensively searched the minimum energy reaction paths of CO2RR on each TM@In2Se3 (TM = Ni, Pd, Rh, Zr, Nb and Re) catalyst (Supplementary Fig. 20). Table 2 summarizes the effects of the polarization switching on the CO2RR, from the aspects of potential determining steps, limiting potential (Ul), and final product.

Table 2 Potential determining steps (PDS), limiting potentials (Ul, V vs RHE), and final product of CO2RR with TM@In2Se3 catalysts.

Depending on the effects of the reversible polarization, the ferroelectric catalysts for CO2RR can be classified into three categories: (i) TM@In2Se3 (TM = Ni, Pd, Rh), for which the polarization switching changes the reaction barriers, but not the reaction paths and the intermediate/final products; (ii) Zr@In2Se3, for which the ferroelectric switching can reactivate the stuck CO2 reduction; (iii) TM@In2Se3 (TM = Nb, Re), for which all the reaction paths, reaction barriers, and final products are polarization dependent. This classification is well consistent with d band center analysis (Fig. 1c), as well as the binding energy, charge transfer, and activation degree of CO2 as listed in Table 1, which indicates clear underlying relationship among these material properties and catalytic performance.

Tunable limiting potential of CO2RR (on TM@In2Se3, TM = Ni, Pd, Rh)

Since the activation degree of CO2 on the ferroelectric surface depends on polarization direction due to the adjusted empty and occupied d-orbitals (shifted d band center), the reaction barrier of the CO2 hydrogenation process is also affected. Taking the Rh@In2Se3 catalyst as an example (see Fig. 4a), though the configurations of activated CO2 (with C atom connecting with Rh atom) under different polarization directions are the same, the energy barriers are obviously different: the reaction of CO2* →OCOH* is downhill on Rh@P ↑ -In2Se3, but uphill on Rh@P ↓ -In2Se3. In the subsequent hydrogenation steps, the polarization reversal does not change the most favourable path of CO2RR, both following the paths of CO2* →OCOH* →CO* →CHO* →CH2O* →CH2OH* →CH3OH* →OH* →H2O*, with the final product of CH4. However, the reaction barrier for each step, the potential-determining step (PDS), and the limiting potential are all different and polarization dependent. On Rh@P ↓ -In2Se3, the PDS of CO2 reduction is CO* →CHO* with a limiting potential of −0.43 V vs. RHE; in contrast, the corresponding PDS on Rh@P ↑ -In2Se3 is the protonation of CHO* to CH2O* with the estimated Ul of −0.26 V vs. RHE. Obviously, the reaction barrier of CO2 reduction is polarization tunable via ferroelectric switching. Notably, the limiting potential (−0.26 eV vs. RHE) is better than most reported CO2RR catalysts (listed in Supplementary Table 6). Therefore, Rh@In2Se3 is potentially an efficient electrochemical catalyst for CO2RR with controllable performance. Similar phenomena of polarization dependent catalysis also exist in Ni@In2Se3 and Pd@In2Se3 (Table 2 and Supplementary Fig. 21). The relatively weaker but still obvious ferroelectric dependent CO2 reduction relates with the moderately polarization dependent d band center (−1.98~−1.16 eV at P ↓ ; −4.13~−3.16 eV at P ↑ , Fig. 1c) and modest activation degree of CO2, as indicated by the moderate CO2 binding energies (Eb-CO2, 0.79-1.39 eV) and mild charge gained by the adsorbed CO2 molecule (QCO2, 0.26-0.45e) (Table 1).

Fig. 4: CO2RR paths on Rh@In2Se3 and Zr@In2Se3.
figure 4

The free-energy profile for the CO2 electrochemical reduction reactions along the minimum energy path at 0 V (vs. RHE) on a Rh@In2Se3 and b Zr@In2Se3. The insets show the optimized configurations of the intermediates. The pink (light blue) shaded area indicates the catalytic reaction when the polarization is pointing upwards (downwards). This color scheme is also used in Fig. 5.

Reactivating stuck catalytic reaction of CO2RR (on Zr@In2Se3)

Zr@P ↓ -In2Se3 cannot activate CO2 molecules, it is thus unsuitable for CO2 reduction. In contrast, CO2 can be activated on Zr@P ↑ -In2Se3 with two oxygen atoms bridging over Zr atom (Supplementary Fig. 18i, j). With more protons, it follows the hydrogenation steps of CO2RR as * + CO2 + H + +e → OCHO* and OCHO* +H + +e → HOCHO*; both of these steps are downhill, indicating that the reduction can react easily. However, further hydrogenation step is prevented by the strong binding strength of the intermediate HOCHO* (∆G = 2.33 eV, Fig. 4b) due to significant electron transfer and over-activations (1.13 e between CO2 and catalyst, 126.2° of OCO as shown in Table 1). Based on the Sabatier principles, such strong binding energies make further reduction only possible at very negative potentials. Further hygenation is thus prehibitted.

Since polarization can modulate interaction strength as indicated above, it is expected that the binding strength of the intermediate HOCHO* can be weakened when the polarization direction of In2Se3 is reversed, so that the catalytic reduction can proceed. Indeed, we found that when polarization direction is reversed from up to down at this step, the barrier decreases (∆G = 0.61 eV for the reduction of HOCHO* to CHO* on Zr@P ↓ -In2Se3). More importantly, HOCHO* on the Zr@P ↓ -In2Se3 is merely 40 meV higher than that on Zr@P ↑ -In2Se3, which indicates that the polarization switching is relatively easy to achieve. The conclusion holds true under dilute Zr concentrations, while the local ferroelectric switching around HOCHO* is prohibited due to the higher energy barrier (see Supplementary Fig. 22). Additionally, according to the subsequent hydrogenation steps, the Ul for Zr@P ↓ -In2Se3 is −0.77 V vs. RHE, which is significantly lower than that for Zr@P ↑ -In2Se3 (−2.33 V vs. RHE). Moreover, the H2O* at Zr@P ↓ -In2Se3 could be removed spontaneously, in contrast to the 0.36 eV energy barrier for the dehydration process on Zr@P ↑ -In2Se3.

Therefore, though CO2 molecule cannot be activated on Zr@P ↓ -In2Se3 for reduction, the polarization reversal can reactivate the catalytic reduction on Zr@P ↓ -In2Se3 with a reasonable limiting potential. During the hydrogenation process, appropriate polarization switching could either reactivate the CO2 reduction or accelerate the reaction rate by reducing the reaction barriers.

Tuning CO2RR path and final product (on Nb@In2Se3 and Re@In2Se3)

In contrast to the aforementioned TM@In2Se3 (TM = Ni, Pd, Rh), the activations of CO2 molecule are much deeper on Nb@In2Se3 and Re@In2Se3, as evidenced by the smaller bond angle OCO (Table 1), much larger Eb-CO2 values (2.02-2.54 eV), and more pronounced charge transfer between CO2 and the catalyst (QCO2, 0.61-0.92e). The huge differences of d band center (−2.52~−2.52 eV at P ↓ ; −7.03~−6.72 eV at P ↑ , see Fig. 1c) render the catalytic reaction more sensitive to the polarization switching, leading to different reaction paths and even different final products according to the Sebastian principles.

Indeed, the polarization switching in the catalysts of Nb@In2Se3 and Re@In2Se3 could partially or even totally alter the CO2RR path, thus leading to the different final products. As illustrated in Fig. 5a, for CO2RR on Re@P ↓ -In2Se3 catalyst, the minimum energy path is CO2* →OCHO* →OCH2O* →HOCH2O* →O* →OH* →H2O*, the final product is methanol, and the PDS is OH* →H2O* with Ul of 0.65 eV; In sharp contrast, the reaction paths on Re@P ↑ -In2Se3 is CO2* →OCOH* →CO* →COH* →C* →CH* →CH2* →CH3* →CH4*, the final product is methane, and the PDS is COH* →C* with Ul of 0.85 eV. Similar CO2RR phenomena were also observed on Nb@In2Se3 catalyst. The most favourable path on Nb@P ↓ -In2Se3 surface is CO2* →OCHO* →OCH2O* →HOCH2O* →O* →OH* →H2O*, producing methanol with Ul of 0.64 eV; in comparison, the CO2* →OCHO* →HOCHO* path on Nb@P ↑ -In2Se3 produces methanoic acid with Ul of 0.58 eV (Fig. 5b).

Fig. 5: CO2RR paths on Re@In2Se3 and Nb@In2Se3.
figure 5

Free energy profile for CO2 electrochemical reduction reactions along the minimum energy path at 0 V (vs. RHE) on a Re@In2Se3 and b Nb@In2Se3. The insets show the optimized configurations of the intermediates.

Note that selective generation of desired chemical fuel products has been pursued in the development of electrochemical CO2RR catalyst over decades5, our strategies via feasible ferroelectric switching can achieve the goal without complex and high-cost steps that use different catalysts or electrolytes to produce desired fuels67,68,69,70. As shown in Fig. 6, our proposed ferroelectric SACs base on experimentally available In2Se3 and unique polarization dependent catalytic performance thus open an avenue for controllable CO2 reduction.

Fig. 6: Schematic diagram of ferroelectric controllable CO2RR.
figure 6

Here, metal anchored α-In2Se3 monolayer is placed between the electrodes to achieve ferroelectric switching and controllable catalysis, tuned by the reversal of the bias direction.


In conclusion, by comprehensive DFT computations, we have screened out six TM@In2Se3 (TM = Ni, Nb, Pd, Re, Rh and Zr) as effective ferroelectric catalysts for electrochemical CO2RR. We found that the polarization reversal of ferroelectric In2Se3 monolayer can adjust the empty/occupied d orbitals of adsorbed TM atom and the energy of the d electrons, thereby tuning the catalytic performance for CO2RR, including the degree of CO2 activation, limiting potential, reaction path, and final product, even reactivating stuck catalytic reduction. Especially, the Rh@P ↑ -In2Se3 catalyst is indentified as a highly efficient electrochemical catalyst for CO2RR due to the fairly low limiting potential (<0.5 V)71. Moreover, the Re@In2Se3 and Nb@In2Se3 catalysts can realize selective generation of different desired products with the same catalyst via polarization switching, which opens a path toward ferroelectric controllable catalysis. These SACs based on 2D ferroelectric materials hold great promise for improving catalytic activity and selectivity for electrochemical CO2RR.


Computational details

Our density functional theory (DFT) computations are performed with the Vienna ab initio simulation package (VASP) code72,73. The spin-polarized generalized gradient approximation (GGA) in the form of Perdew − Burke − Ernzerhof (PBE) treats the exchange − correlation interactions, while the frozen-core projector augmented wave (PAW) approximation describes the interaction between the ion and electron74,75,76. The van der Waals (vdW) interactions are described with the DFT-D3 method in Grimme’s scheme77. The models with one TM atom anchored 2 × 2 In2Se3 are built as the potential catalysts for the simulations, since it corresponds to the coverage of 25% with the optimal adsorption energies, as shown in Supplementary Fig. 23. Four TM atoms uniformly distributed on the hexagonal centers of 4 × 4 In2Se3 supercell act as the catalytic active sites. More than 20 Å vacuum space perpendicular to the surface is added to avoid interactions between periodic slabs. The dipole correction is taken into account for all the asymmetric structures78. The 2D Brillouin sampling is presented with a 4 × 4 × 1 gamma-centered Monkhorst-pack k-mesh. The cutoff energy for plane-wave basis sets is 500 eV. The convergence thresholds for force and total energy are 10−2 eV/Å and 10−5 eV, respectively. Moreover, the Gibbs free energy calculation is operated with computational hydrogen electrode (CHE) model79, and the solvent effect is considered with the implicit solvent model implemented in VASPsol27,61. The site-specific charge differences are obtained using Bader analysis. Besides the standard PBE calculations and the models of TM@2 × 2 In2Se3, we also reexamined the chemical reactions by using larger supercells (TM@4 × 4 or 6 × 6 In2Se3, Supplementary Fig. 24) or different exchange functions (PBE + U or RPBE; Supplementary Figs. 25 and 26). All these approaches lead to the same conclusion about the ferroelectric controlled CO2 reduction, with the key numerical results only slightly different, indicating the robustness of our conclusion. More details of the simulations can be found in the Supplementary Information.