Introduction

Electrocatalytic nitrogen reduction reaction (eNRR) has garnered significant attention as a crucial process for ammonia production and related applications1,2,3. In contrast to the main industrial Haber-Bosch process for producing ammonia under high temperature and pressure, the eNRR operates efficiently under milder conditions and consumes less carbon2,3,4. In this catalytic process, chemical changes take place through successive hydrogenation reactions to break the N≡N triple bond, including the distal and alternating pathways3. In the eNRR process, it is crucial that the catalyst must be able to chemically activate N2, an inert gas that does not adsorb on the surfaces of most materials5,6,7. It is therefore highly significant to search for efficient eNRR catalysts and explore related catalytic mechanisms.

In pursuit of potential catalysts, two noble metals, Ru and Re, have been identified as promising catalysts for eNRR among the transition metals8,9. However, these systems not only exhibit higher limiting potentials along the reaction trajectories but also show poor selectivity compared to the competing hydrogen evolution reaction (HER) process. Single-atom catalysts (SACs) exhibit advantage due to their cost-effectiveness and versatile tunability, and have attracted significant attention since the conceptual inception10,11,12. In the context of eNRR, localized electronic states of the SACs near the Fermi level facilitate N2 adsorption and promote completion of the reaction process. For example, it was proposed that FeN3-embedded graphene can be a catalyst for eNRR based on first-principles calculations13. Subsequently, g-C3N4 and defective boron nitride monolayers have also been predicted as eNRR catalysts within the context of SACs14,15,16. In addition, catalytic mechanisms involving diatomic or triatomic configurations have further been proposed for eNRR17,18,19,20,21,22,23. Despite the advantages of discrete atoms for precise control of the active sites, it is still demanding to discover new types of catalysts with high-density (or high-loading) and highly efficient active sites. However, achieving high loadings of SACs represents a new challenge, because such SACs in close proximity can lead to interatomic couplings and cluster formation24. Furthermore, ensuring catalytic stability of the anchored SACs presents another challenging aspect25,26.

Based on these earlier developments, we infer that materials featuring both naturally dispersed atoms and localized electronic states may serve as promising candidates for catalyzing eNRR. These considerations direct us to the actively investigated Kagome materials. In this new class of quantum materials, their two-dimensional (2D) Kagome lattices possess a triangular structure of the transition metal atoms. Furthermore, these lattices exhibit geometric frustrations, which lead to a rich spectrum of unconventional electronic properties, including nontrivial topology, van Hove singularities, and, most notably, flat electronic bands27,28,29,30,31. Specifically, Co3Sn2S2 exhibits giant anomalous Hall effect and Weyl topology32,33, and Fe3Sn2 hosts Dirac fermions34. In particular, the FeSn35,36,37,38,39, CoSn40,41,42, and FeGe systems feature flat electronic bands and nontrivial band topology43,44,45. Beyond such physical properties, the catalytic activity of Co3Sn2S2 in oxygen evolution reaction (OER) has also been investigated46, emphasizing the pivotal role of the topological surface states in enhancing the catalytic performance47,48.

In the present study, our primary emphasis is to investigate the potential role of the flat band characteristics of Kagome materials in facilitating the eNRR by integrating the flat band physics with SACs. Through such a conceptual integration, we aim to identify efficient eNRR catalysts within the Kagome materials, while fundamentally overcoming the severe limitations encountered by the artificially anchored single- or few-atom catalysts. Specifically, we employ first-principles calculations to demonstrate that the appropriately terminated surface of the Kagome metal FeSn can serve as a highly efficient catalytic platform for eNRR, featuring naturally abundant active sites, high catalytic stability and high chemical selectivity. We first identify the naturally occurring 2D Kagome surface through cleavage, and such a termination is accompanied by surface flat electronic bands stemming from the Fe-dz² and dxz/dyz orbitals. We further demonstrate that N2 can strongly chemisorb atop an Fe atom in the Kagome surface layer. Moreover, the limiting potential along the entire reaction pathway is calculated to be only 0.31 V within the distal mechanism, highlighting the superior eNRR catalytic performance3. Our analysis of the electronic structure variations and magnetic properties underscores the pivotal role of the flat bands in the catalytic activity. The mutual independence between neighboring reactive sites also ensures an exceptionally high 25% atomic utilization within the Kagome layer. In addition, the high selectivity of eNRR over potential competing HER is validated through our detailed studies. We also generalize our design principle for efficient eNRR catalysts to the isostructural CoSn and FeGe Kagome systems, which further demonstrate the crucial role of the flat electronic bands. Collectively, this work not only effectively enhances the functionalities of Kagome materials for important practical applications but also integrates flat band physics with single-atom catalysis, offering new opportunities in catalyst design.

Results

Atomic and electronic structures of Kagome-terminated FeSn

We start by investigating the feasibility of exposing the Kagome termination in FeSn. As shown in Fig. 1a, b, FeSn consists of alternating layers of Fe3Sn and Sn2, with the Fe atoms in the Fe3Sn layer forming a Kagome structure. The interlayer interaction is limited to Fe3Sn–Sn2, thus, cutting FeSn along the (001) direction will result in the coexistence of both Sn2 and Fe3Sn surface terminations, as confirmed experimentally35,36,37,38,39. Furthermore, a Kagome layer is separated from the Sn layer by approximately an interlayer distance of 2.2 Å, suggesting relatively weak interlayer coupling. Thus, FeSn provides a favorable platform for investigating the catalytic properties on a quasi-two-dimensional Kagome lattice. Similar findings are also applicable to the isostructural Kagome systems of FeGe and CoSn42,44. In contrast, several other widely studied Kagome systems, such as Co3Sn2S2 and AV3Sb5, cannot be readily prepared to expose their Kagome terminations32,49.

Fig. 1: Atomic and electronic structures of Kagome-terminated FeSn.
figure 1

a Crystal structure of FeSn. b Top view of the Kagome surface. c Orbital-resolved band structure and d projected density of states (PDOS) for the Kagome termination of FeSn. The highlighted regions by two black lines in (c) and two red lines in (d) mark the flat bands (FBs).

Full size image

Next, we turn our attention to the electronic structure of the Kagome-terminated FeSn. In Fig. 1c, d, we present the orbital-resolved band structure and projected density of states (PDOS) for FeSn with the Fe3Sn Kagome termination, respectively. Within the energy range of −0.3 to −0.1 eV, discernible flat bands are observed, primarily originating from the contributions of the Fe-dz² and dxz/dyz orbitals. Such flat bands derived from the localized electronic states have been experimentally confirmed35,37,38. We also note that the flat bands exhibit spin-polarized features, which are expected to play a crucial role in breaking the triple bond of the nitrogen molecule13,18,50.

Adsorption of N2 on Kagome-terminated FeSn

Based on density functional theory (DFT) calculations (with calculation details given in Part I of the Supplementary Information (SI)), we investigate the adsorption of a N2 molecule on the Kagome termination of FeSn. First, we examine four distinct adsorption sites: (1) on-top sites of the surface Fe atoms, (2) bridge sites of the Fe atoms, (3) hollow sites of the Fe atoms, and (4) on-top sites of the surface Sn atoms, as shown in Supplementary Fig. S1. Our calculations indicate that N2 can strongly chemisorb on the on-top site of a surface Fe atom, with an adsorption energy of −0.74 eV, while showing instability at the other three sites. In Fig. 2a, we present the adsorption configuration along with the distribution of charge density differences. We see that N2 can adsorb on the adsorption site in an end-on configuration, bonded to Fe, representing a widely studied back-donation adsorption mechanism51. Our detailed Bader charge analysis further reveals a significant electron transfer of 0.52 e− from FeSn to N2.

Fig. 2: Atomic structure and electronic property analysis of N₂ adsorption on Kagome-terminated FeSn.
figure 2

a Adsorption configurations of N2 along with distribution of charge density differences from side view on the Kagome surface of FeSn. The blue and yellow areas represent electron depletion and electron accumulation, respectively. b Molecular orbitals of approximately free N2. c PDOS of N2 and the Fe-dz² orbital after N2 adsorption. d PDOS of N2 and the Fe-dxz/dyz orbitals after N2 adsorption.

Full size image

To learn more insights into the adsorption mechanism of N2 on the Kagome surface, we conduct comparative PDOS analyses. First, as a reference system, we consider an isolated N2 molecule sufficiently far from the Kagome surface, showing the highest occupied σ orbital and lowest unoccupied π* orbital, as shown in Fig. 2b. Upon N2 adsorption on the Kagome surface, we observe hybridization between the Fe-dz² orbital and σ orbital of the N2 molecule (Fig. 2c), whereas the Fe-dxz/dyz orbitals display significant hybridization with the occupied π orbital and unoccupied high-energy π* orbital (Fig. 2d), with the coupling resulting from the allowed symmetry of the molecular orbitals. The back-donation mechanism is also confirmed by the partially occupied d-π* orbital. While these analyses are consistent with chemical principles, our objective is to investigate the physical basis underlying the enhanced N2 adsorption on the Kagome surface. To do so, we investigate the PDOS changes during the adsorption process. The comparison is made between the PDOS at the adsorption site and that at an Fe site distant from the adsorption site, as defined in Supplementary Fig. S1. From Fig. 3a, we observe significant changes in the PDOS of the dz² orbital, which predominantly contributes to the flat bands. The dxz/dyz orbitals also show notable changes in the PDOS near the Fermi level, whereas the in-plane dxy/dx²−y² components remain virtually unchanged throughout the process, as displayed in Supplementary Figs. S2 and Fig. S3. These observations underscore the pivotal role played by the flat bands during the N2 adsorption process on FeSn.

Fig. 3: PDOS analysis for various nitrogen species adsorbed on Kagome-terminated FeSn.
figure 3

PDOS of the Fe-dz² orbital of the adsorption site (purple) and distant site (blue) after adsorption of a N2*, b N2H*, c N2H2*, d N*, e NH*, and f NH2*. The definitions of different sites are presented in Supplementary Fig. S1.

Full size image

Subsequent steps of the ammonia production

To complete the ammonia production cycle, we further investigate the process after N2 activation. We observe that subsequent intermediates of the eNRR do not adsorb on the on-top site of Fe but instead occupy the hollow or bridge sites of Fe, as shown in Fig. 4. We conjecture that after N2 activation, the adsorption of subsequent intermediates depends significantly on the charge donation capability of the catalysts. In extensive prior studies, it has been consistently observed that the adsorption of the N2H* intermediate typically represents the rate-limiting step in the whole eNRR process8,9. In our present study, when adsorbed at the hollow site, the N atoms maximize their interaction with the host, thereby significantly lowering the associated energy barrier to be comparable to that in the well-studied eNRR process on a Ru (0001) surface8. Similar observations are evident in the charge density difference diagrams for the adsorption configurations of the subsequent intermediates, as shown in Supplementary Fig. S4.

Fig. 4: Free energy diagram for the distal pathway of eNRR on Kagome-terminated FeSn.
figure 4

Free energy diagram as well as the optimized geometric configurations of different intermediates in the eNRR process, with an overall limiting potential of 0.31 V. In Figs. 4 and 5, the different atoms are indicated by the same color codes as in Fig. 1.

Full size image

In analogy to the analysis of N2 adsorption discussed above, we have further monitored the changes in PDOS for the subsequent steps of the ammonia production. We find that the PDOS contributions from the dz² orbital exhibit the most significant variations in each subsequent reaction step, and notable changes in the PDOS of the dxz/dyz orbitals are also observed, highlighting the critical involvement of the flat bands, as shown in Fig. 3b–f and Supplementary Figs. S2 and S3. In Fig. 4, we illustrate the free energy barrier diagram for the distal pathway of eNRR. The overall limiting potential is 0.31 V, which surpasses the performance of the most efficient noble metal surfaces for eNRR activity8. As for the alternating mechanism, the limiting potential is higher (0.98 V), indicating that the distal mechanism dominates, and the corresponding Gibbs free energy diagram is presented in Supplementary Fig. S5. Based on these analyses, we conclude that the flat bands of the Kagome surface play a critical role throughout the entire eNRR process, as anticipated.

Robustness of the catalytic activity of Kagome-terminated FeSn

To elucidate the advantages of the natural Kagome FeSn system compared to artificially anchored single or few-atom catalysts, we explore the catalytic activity of the Kagome surface at varying adsorbate coverage. Considering the equivalence of the Fe atoms on the Kagome surface, we focus on the 2 × 2 supercell configuration. Here, each Kagome layer contains 12 Fe atoms organized into triangular units, with 4 such units per supercell, as illustrated in Fig. 5a. To investigate the impact of the adsorbate coverage and validate the independence between the neighboring active sites, we vary the number of N* species (ranging from 0 to 3) occupying the reaction units, and assess the adsorption capacity of the remaining vacant sites for N2. As shown in Fig. 5b–d, when pre-adsorbing 0 to 3 N* species, the corresponding adsorption energies ΔEN₂ are −0.74, −0.66, −0.66, and −0.62 eV, respectively, confirming that these individual sites indeed exhibit sufficient independence in their reactivity. Overall, the atomic utilization efficiency of the monolayer reaches 25% (namely, 4 active units among the 16 atoms). Compared to the atomic percentage (at%) concept within the SACs framework, we conclude that the present system exhibits superior performance as an eNRR catalyst.

Fig. 5: N₂ Adsorption on the Kagome surface with different N* coverage.
figure 5

a Four triangular units defined in the 2 × 2 supercell of the Kagome Fe3Sn layer. Configurations of N2 adsorption on the Kagome surface pre-adsorbed with b one N*, c two N*, and d three N*.

Full size image

Going further into the underlying mechanism, we explore the potential impact of spin polarization on the reaction dynamics. As hypothesized, according to Hund’s rule, electrons occupying the empty π* orbital of N2 tend to favor parallel spin alignment, suggesting that spin-polarized electrons from the catalyst may lower the reaction barriers. In our DFT calculations, the Fe atoms in the surface Kagome layer carry spin-down magnetic moments, as shown in Supplementary Fig. S6, while the flat bands depicted in Fig. 1d exhibit spin-up polarization. We also present the magnetic moment carried by each intermediate species after its adsorption in Supplementary Table S2. Although not prominently pronounced, each intermediate species shows a spin-up magnetic moment, where N2* carries a magnetic moment of +0.1 μB, and N2H* carries a magnetic moment of +0.17 μB. These results not only underscore that the flat band plays a predominant role in the interaction with the adsorbate but also further indicate the preference for spin-polarized electrons in these reaction pathways.

Here, it is worthwhile to note that solvent effects and electrode potential effects can both influence the calculated results for catalytic processes52,53,54. To investigate these effects on the catalytic activity of FeSn, we conduct additional calculations by incorporating solvent effects using an implicit solvent model and constant potential calculations using the grand canonical ensemble DFT (GCE-DFT) approach55,56. The corresponding free energy changes for the eNRR catalytic process are presented in Supplementary Tables S3 and S4. Although these factors slightly modify the free energy changes for the steps of the eNRR process, the main conclusions remain unchanged. Moreover, neither the solvent effects nor the electrode potential effects alter the underlying catalytic mechanisms of the Kagome surface of FeSn, particularly the critical role of the flat bands in the eNRR process. For the HER process, these effects are shown to be negligible. In addition, eNRR activity is influenced by both the reaction kinetics and thermodynamics. To obtain the reaction kinetics for electrocatalytic processes, explicit solvent modeling should be employed, such as by explicitly incorporating water layers52,53, which would significantly increase the complexity and computational cost. Instead, many previous studies of electrocatalysis have suggested that the computational framework based on the computational hydrogen electrode (CHE) model, as adopted in this study, can provide reasonably accurate predictions of the reaction trends57,58.

To assess the selectivity of eNRR over HER on FeSn, we show the Faradaic efficiency, as employed in previous literature16. The Faradaic efficiency of eNRR can be expressed as

$${f}_{\mathrm{NRR}}=\frac{100 \% }{1+{e}^{\frac{-\delta G}{{k}_{B}T}}},$$
(1)

where δG denotes the Gibbs free energy difference between the potential-determining steps of HER and eNRR at 0 V vs. the standard hydrogen electrode (SHE), kB represents the Boltzmann constant, and T is taken as the room temperature and is set to be 298.15 K. For a catalytic reaction, among all elementary steps, the step with the highest ΔG defines the potential-determining step, and this ΔG value corresponds to the limiting potential (vs. SHE). For HER, the hydrogen adsorption energy is calculated to be −0.92 eV, and thereby the limiting potential is determined to be 0.74 eV; for eNRR, the limiting potential is 0.31 eV. The calculated value of \({f}_{\mathrm{NRR}}\) is essentially 100%, demonstrating the high selectivity of FeSn toward eNRR. A comparison of the volcano plots for HER and eNRR on transition metals indicates that noble-metal-based eNRR catalysts intrinsically suffer from low selectivity, as the HER pathway invariably dominates on their surfaces5,7,8,16,18. In contrast, the present Kagome system exhibits markedly superior performance, with the limiting potential for eNRR being significantly lower than that for HER.

Catalytic activity of other Kagome materials for eNRR

In order to substantiate the general applicability of the significant conclusion regarding the catalytic activity of Kagome flat bands in eNRR, we conduct related studies of the two isostructural Kagome materials CoSn and FeGe. We find that both CoSn and FeGe exhibit N2 adsorption capabilities, with the adsorption energies of −0.81 and −0.50 eV, respectively. Further investigations of the overall eNRR process on their Kagome terminations show that the limiting potential barriers are 0.70 and 0.68 eV through the distal mechanism, respectively. The corresponding free energy barrier diagrams are displayed in Supplementary Figs. S9 and S11, and the optimized geometric configurations of diverse intermediates along the distal pathway are shown in Supplementary Figs. S10 and S12. As expected, due to differences in electron counting and magnetic states, CoSn and FeSn show variations in their composition of the flat bands near the Fermi level. In the energy range of −0.3 to 0 eV, the dxz/dyz orbitals make predominant contributions to the flat bands of CoSn, whereas the dxy/dx²−y² orbitals also contribute to the flat bands near the Fermi level, as shown in Supplementary Fig. S13. Upon N2 adsorption, we observe significant variations in the PDOS associated with the dxz/dyz and dxy/dx²−y² orbitals, whereas the dz² orbital, which does not significantly contribute to the flat bands, shows relatively minimal PDOS changes near the Fermi level (see Supplementary Figs. S14–S16). These findings further confirm that the flat bands can consistently dominate the overall chemical process, rather than being restricted to certain specific orbitals. Our results not only underscore the significant role of flat bands but also provide constructive guidance for the targeted design of catalysts based on Kagome structures.

Discussion

In summary, we have identified that FeSn, a Kagome material, can function as a highly efficient catalyst for eNRR, with naturally abundant active sites, high stability and high selectivity. We have shown that the 2D Kagome surface is accompanied by the presence of surface flat electronic bands contributed by the Fe-dz² and dxz/dyz orbitals, resulting in chemisorption of N2. The calculated limiting potential for the whole process is only 0.31 V through the distal pathway, indicating the system’s superior eNRR catalytic activity. The underlying mechanism for this enhanced catalytic activity is attributed to the presence of flat bands. The mutual independence between neighboring reactive sites further ensures an exceptionally high 25% atomic utilization within the Kagome layer, with each active site possessing high selectivity of eNRR over competing HER. We have also generalized our design principle for efficient eNRR catalysts to the isostructural CoSn and FeGe Kagome systems, which further demonstrate the crucial role of the flat electronic bands. Overall, this work not only effectively enhances the functionalities of Kagome materials for important practical applications, but also integrates flat band physics with single-atom catalysis, offering new opportunities in catalyst design.

Methods

Density functional theory (DFT) calculations

We employed the Vienna Ab initio Simulation Package (VASP) within density functional theory (DFT) for first-principles calculations59,60,61. The Perdew–Burke–Ernzerhof (PBE) generalized gradient approximation (GGA) functional was adopted, incorporating spin polarization and DFT-D2 for van der Waals interactions62,63. 2 × 2 × 2 supercells were used for the antiferromagnetic FeSn and FeGe, and a 2 × 2 × 4 supercell for the paramagnetic CoSn. The convergence criteria were set at −0.01 eV/Å for force and 10−5 eV for energy. A 20-Å vacuum layer was applied to prevent interaction between different images. During optimization, the top four atomic layers were left free to relax, while the remaining four layers were fixed. Bader charge analysis was employed to investigate the charge transfer processes64.

The adsorption energy of N₂ is defined as:

$$\Delta \,{{\rm{E}}}_{{{\rm{N}}}_{2}}\,=\,{{\rm{E}}}_{\mathrm{sub}+{{\rm{N}}}_{2}}\,-\,{{\rm{E}}}_{\mathrm{sub}}\,-\,{{\rm{E}}}_{{{\rm{N}}}_{2}},$$
(2)

where Esub+N₂, Esub, and EN₂ are the total energies of the combined system, substrate, and an isolated N2, respectively. The change in Gibbs free energy, ΔG, for each elementary step was determined within the computational hydrogen electrode (CHE) framework65, given by:

$$\Delta \,{\rm{G}}\,=\Delta \,{\rm{E}}\,+\,\Delta \,{{\rm{E}}}_{\mathrm{ZPE}}\,-\,{\rm{T}}\,\varDelta \,{\rm{S}}.$$
(3)

Here, ΔE represents the energy difference between the product and reactants; ΔEZPE and ΔS denote the changes in zero-point energy and entropy, respectively, which are determined from vibrational frequency calculations; T is the temperature. Detailed descriptions are presented in Part I of the SI.

Solvent effects were incorporated into the DFT calculations by utilizing the VASPsol code66,67, implemented in VASP. VASPsol employs an implicit solvent model, which describes the solvent environment through continuous dielectric properties rather than by explicitly representing individual solvent molecules. This approach minimizes the computational overhead associated with explicitly modeling solvent molecules, while still capturing the essential physics of solvent–solute interactions.

The grand canonical ensemble density functional theory (GCE-DFT) constant potential calculations55,56 were carried out within the VASP software. The electrode potential was fixed by adjusting the total number of electrons in the system via the “NELECT” parameter in VASP. The absolute electrode potential (Φe) is determined by the following equation:

$${\Phi }_{e}={\Phi }_{w}-{E}_{F},$$
(4)

where Φw represents the work function of the system, and EF is the Fermi level. The absolute electrode potential relative to the SHE, denoted as USHE, is further defined as:

$$e{U}_{SHE}={\Phi }_{e}-{\Phi }_{SHE}.$$
(5)

Here, ΦSHE has been experimentally determined to be approximately 4.44 eV, and e is the elementary charge. Then, the grand potential Ω is defined as:

$$\Omega ={E}_{\mathrm{tot}}+{\Phi }_{e}{N}_{e},$$
(6)

where Etot is the total energy of the system, and Ne is the total number of electrons.