From double-atom catalysts to single-cluster catalysts: A new frontier in heterogeneous catalysis

Abstract

In recent years, atomically dispersed metal catalysts with well-defined structures have attracted great interest in heterogeneous catalysis due to their high atomic utilization efficiency, activity, stability, and selectivity. The rapid development of single-atom catalysts (SACs) has simultaneously stimulated the emergence of diatomic catalysts (DACs) and single cluster catalysts (SCCs). Compared with SACs, DACs, and SCCs possess higher metal loading and more structurally flexible active sites, which provide great potential for achieving higher catalytic performance. DACs and SCCs have become a new field of heterogeneous catalysis. In this review, we first focus on the latest developments of DACs/SCCs, including synthesis methods and characterization approaches including experimental and theoretical tools. In addition, the relationship between structure and catalytic performance of DACs/SCCs are well discussed, including the effect of supports, synergistic metal atoms, and coordination environment. At last, the similarities and differences between SACs and DACs/SCCs are systematically summarized and analyzed.

INTRODUCTION

Since the pioneering work of Fe-based single-atomic site catalyst (denoted Fe₁/FeO_x) was reported by Zhang in 2011, single-atom catalysts (SACs) have attracted widespread research attention and have become a frontier in the field of catalysis.^[1] After that SACs have developed rapidly because of the following advantages. (1) SACs with dispersed metal atoms on the support can achieve 100% atomic utilization,^[2] rationally using metal resources and adding the number of active centers.^[3] (2) Uniformity active centers of SACs make them possess high selectivity.^[4–7] (3) The unique electronic structure and unsaturated coordination environment of the active center can improve the catalytic activity.^[8–12] All of these properties enable SACs to possess the advantages of heterogeneous catalysts and homogeneous catalysts and shows good stability and high activity in various reactions,^[13–15] such as electrocatalysis, hydrogenation, and oxidation reactions. However, SACs still face some limitations. First, the active site of SACs is an isolated atom, which only provides a single adsorption site for the activated reactant. Nevertheless, most catalytic reactions usually involve co-adsorption of multiple reactants and the formation of multiple chemical bonds on the catalyst surface by intermediate molecules.^[16–20] Second, in order to prevent sintering and agglomeration during the reaction derived from high surface energy of single atoms, many experimental studies of single atom catalysts are devoted to improving the dispersion. The distance between adjacent sites are relatively long (>>1 nm), so that the loading of single metal atoms is usually not high. However, low metal loading results in the unsatisfactory quality activity of the overall catalyst.^[21–24] These limitations have severely restricted the development and application of SACs.

In recent years, researchers have gradually designed and constructed atomic-level dispersion catalysts with multiple atomic active sites and higher metal loadings. Thus, double-atom catalysts (DACs) and single-cluster catalysts (SCCs) were derived. And the emergence of DACs/SCCs makes up for the deficiencies of SACs. First, DACs and SCCs can anchor more metal atoms at an adsorption site compared with SACs, which indicates that DACs/SCCs have potential to possess higher metal loading than SACs.^[25–27] Second, DACs/SCCs have a discontinuous band structure, which is closely related to the type and the number of metal atoms, and changing atoms in a small cluster may greatly change the electronic structure.^[28] Therefore, the electronic structure of active sites of DACs/SCCs can be easily tuned.^[29] Third, DACs/SCCs provide adjacent unsaturated coordination sites, which can adsorb multiple reactants at the same time.^[30,31] At last, the synergistic effect between the metal atoms can change the adsorption mode of key intermediates and optimize reaction pathways, resulting in significantly improved catalytic activity and selectivity.^[32–35]

Inspired by the obvious advantages above, DACs/SCCs may become a new frontier of atomic-level catalysts in future. Here, we systematically review the recent development of DACs/SCCs (Figure 1). In the second section of this review, we mainly summarize the effective and accurate synthesis method of DACs/SCCs, including the top-down and bottom-up perspectives. The third section demonstrates density functional theory (DFT) calculations and advanced experimental techniques. These techniques provide a detailed understanding on the morphological structure of the catalytically active center at the atomic level, the interaction between metal atoms and substrates, and the corresponding catalytic mechanism. Section 4 discusses the deep insight into the relationship between the structure of DACs/SCCs and their catalytic performance from three aspects, namely support effect, coordination effect, and synergy effect. Section 5 systematically analyzes the similarities and differences between SACs and DACs/SCCs. Finally, the future development opportunities and challenges of DACs/SCCs are proposed.

FIGURE 1. Schematic diagram of the research route starting from multiple-atomic-sites catalysts in this review. A, Ref.[16] Copyright 2019, Nature. B, Ref.[33] Copyright 2018, Nature. C, Ref.[29] Copyright 2019, Wiley-VCH. D, Ref.[36] Copyright 2020, ACS Publications. E, Ref.[30] Copyright 2020, Wiley-VCH. F, Ref.[37] Copyright 2019, Elsevier. G, Ref.[38] Copyright 2020, ACS Publications

SYNTHESIS STRATEGIES

Adjustable synthesis strategies play an important role in preparing uniformly atom-dispersed dimers/trimers. The high-performance catalyst can be prepared by selecting a suitable preparation method to promote the synergistic interaction between the metal sites and the support. So far, the bottom-up strategy and top-down strategy are commonly used to prepare DACs and SCCs.^[39,40]

Bottom-up strategies for DACs/SCCs

Generally, bottom-up metal precursors undergo physical adsorption and chemical reduction to be anchored on the substrates.^[41] To date, several bottom-up strategies for DACs/SCCs have been successfully developed, such as impregnation method, atomic layer deposition (ALD), and electrochemical methods.

The impregnation method is widely used due to its simple operation.^[42] As reported, Wang and co-workers^[43] prepared highly dispersed Fe₂ clusters loaded on mesoporous carbon nitride (mpg-C₃N₄) through an impregnation method of “preselection of precursors.” The pre-selection of the precursor (Fe₂O₄C1₄H₁₀) is essential to the preparation of Fe₂/mpg-C₃N₄. The pyrolysis process was then optimized to completely remove organic ligands from the precursor, while also suppressing aggregated Fe₂ clusters. The obtained Fe₂/mpg-C₃N₄ sample showed superb selectivity of 93% and high conversion of 91% for the epoxidation of trans-stilbene to trans-stilbene oxide. In this regard, Lu et al.^[44] synthesized Fe-NiNC catalyst by a simple two-solvent ion deposition method, which showed outstanding catalytic performance for both OER and ORR. As shown in Figure 2A, nickel-doped polydopamine (Ni-PDA) as the host was dispersed in n-hexane under ultrasonic treatment, and then Fe(NO₃)₃ solution was added dropwise to the suspension. The dissolved Fe ions dispersed into fine droplets and diffused into the Ni-PDA porous surface due to the immiscibility of water and n-hexane. Fe ions were deposited on Ni-doped sites upon adsorption, and Fe-Ni bonds were formed with the annealing treatment. Finally, satisfactory Fe-NiNC-50 hollow spheres were obtained through the proper calcination and chemical leaching operations. Although the impregnation method does not require special equipment and could prepare single-atom catalysts by routine operations, the size distribution of obtained metal catalysts are often non-uniform (from single atoms to nanoparticles).

FIGURE 2. DACs/SCCs synthetic methods via bottom-up strategies. A, Two-solvent ion deposition method[44] Copyright 2020, Elsevier. B, The atomic layer deposition (ALD) method[48] Copyright 2017, Nature. C, In-situ electrochemical deposition method[50] Copyright 2019, ACS Publications. D, Electrochemical potential window strategy[36] Copyright 2020, ACS Publications

The atomic layer deposition (ALD) method is a widely used technique, which deposits metal atoms uniformly on the support with good repeatability.^[45,46] It has been successfully used to construct bimetallic sites.^[47] For instance, Yan and co-workers^[48] successfully prepared the Pt₂ diatomic catalyst (Pt₂/Graphene) via a two-step ALD process. As shown in Figure 2B, many oxygen-containing functional groups and defects were generated on the original graphene nanosheets after acid oxidation and high-temperature thermal reduction treatment. The first ALD cycle was performed at 250℃ to obtain Pt₁ single-atomic catalyst (Pt₁/Graphene) by alternately exposing MeCpPtMe₃ and molecules O₂ on the graphene surface. Then, the reaction temperature was adjusted to 150℃ during the second ALD cycle. MeCpPtMe₃ was only adsorbed on a single atom of Pt₁ to obtain Pt₂ diatomic atomic catalyst (Pt₂/Graphene). Moreover, Sun et al.^[49] successfully prepared the high-quality Pt-Ru heteronuclear dimers on nitrogen-doped carbon nanotubes (NCNTs) by a two-step ALD approach. Although the precise and controllable ALD technology is widely used to prepare high-dispersion and high-loading atomic-level catalysts, its high cost is not suitable for industrial application.

At present, electrochemical methods for the preparation of high purity DACs/SCCs have been developed, such as in situ electrochemical deposition and electrochemical potential window strategy. For instance, Hu et al.^[50] revealed that Co-Fe double-atom catalyst (Co-Fe-N-C) was derived from a monoatomic Co pre-catalyst by in-situ electrochemical deposition. Under positive potentials, the iron-containing alkaline electrolyte [Fe(OH)₄]^– tends to be adsorbed on the Co species of the pre-catalyst (Co-N-C). After electrochemical activation treatment to atomically disperse Co species onto Fe-containing alkaline electrolyte, a Co-Fe-N-C catalyst was formed and shown in Figure 2C. Compared with Co-N-C, the incorporation of iron was crucial for the enhanced efficiency of Co-Fe-N-C, and its activity increased with the amount of Fe. Very recently, Li et al.^[36] reported an electrochemical potential window strategy for the preparation of Pt₃ and Ni₃ SCCs anchored on graphdiyne. The principle of this strategy was divided into two steps (see Figure 2D). The metal atoms deposited on the substrate by using conventional techniques without precise control, and then there can be a sufficiently large electrochemical potential window under the application of voltage. In this window, all metals were oxidized and leached out except for the firmly bonded SA/SC, which could remove the target metal in other form on the substrate. Although the electrochemical synthesis method has less energy consumption, it is still not suitable for large scale production.

Top-down strategies for DACs/SCCs

As for the top-down strategy, bulk metallic materials, such as large nanoparticles, metal foam and so on, are often employed as precursor, which are calcined to induce thermal migration of metal atoms at temperature more than 800°C and prepare atomically dispersed structure.^[29,51] Compared with bottom-up strategy, the top-down strategy could more efficiently prepare well-defined DACs/SCCs with precise structure.^[52]

Xiong et al.^[37] employed inorganic compounds with different numbers of iron atoms as precursors, which were encapsulated into zeolitic imidazolate framework (ZIF-8) followed by high temperature calcination, forming clusters with different numbers of iron atoms. As shown in Figure 3A, taking the Fe₂ cluster synthesis process as an example, the binuclear Fe₂(CO)₉ compound was encapsulated in the cavity of ZIF-8 in situ to form a Fe₂(CO)₉@ZIF-8 hybrid structure. Subsequently, Fe₂(CO)₉@ZIF-8 in the cavity was decomposed into Fe₂ clusters, while Zn metal nodes in ZIF-8 underwent thermal migration under 800℃. Thus, Fe₂ clusters stably embedded in nitrogen-doped carbon. In this regard, Wu et al.^[53] and Xu et al.^[54] successfully prepared Fe-Co heteronuclear diatomic catalysts. Zn-based triazole-rich energetic MOFs encapsulated with Co and Fe species were used as precursors. After high-temperature pyrolysis, zinc nodes were reduced and volatilized, while MOFs formed a hierarchical porous carbon network. Meanwhile, evenly dispersed Fe and Co target metal ions were uniformly dispersed into the carbon grid. The obtained CoFe@NC catalyst exhibited excellent electrocatalytic performance for oxygen reduction reaction (ORR). Very recently, Xu et al.^[30] used the two-solvent method to encapsulate the trinuclear complex into nitrogen-doped carbon layer (AIST-1) to obtain nano-iron metal clusters, namely Fe^III₂M^II(μ₃-O)(CH₃COO)₆(H₂O)₃ (M = Fe, Zn, Co). In order to suppress the excessive atoms aggregation, Fe^II ions were replaced with other divalent metal ions (e.g., Zn^II/Co^II) to synthesize the isostructural trinuclear complex precursors. This is the “heteroatom modulator approach,” which turns into stable iron dimer after pyrolysis (Figure 3B).

FIGURE 3. Fabrication of DACs/SCCs via top-down strategies. A, Reproduced with permission.[37] Copyright 2019, Elsevier. B, Reproduced with permission.[30] Copyright 2020, Wiley-VCH

ADVANCED METHODS FOR THE CHARACTERIZATION OF DACS AND SCCS

The rapid development of advanced characterization techniques includes experimental techniques (Table 1) and theoretical methods, which has greatly promoted a deep understanding of catalysis and brought the research of catalysis to the atomic scale. Herein, we summarize the characterization techniques of DACs and SCCs.

TABLE 1 The summary of instrumental analysis of catalysts for better catalyst research

Experimental techniques		Application	Ref.
Electron microscopy approach	High Resolution Transmission Electron Microscope (HRTEM)	It can obtain the structure and composition information of particles by combining the results of high-resolution image and energy spectrum analysis. It can also determine which crystal plane belongs to by measuring the spacing of fringes and then comparing it with the standard crystal plane spacing d.	[49,52,64,65]
	Transmission electron microscope (TEM)	Due to the different from the principle of SEM, TEM can be used to observe the internal structure of the sample and samples need to deal with very thin. Through Scanning electron Microscope (SEM) and Transmission electron Microscope (TEM), the surface morphology and microstructure of the catalytic material were observed.	[35,64,65]
	Scanning electron microscope (SEM)	It has advantages of wide magnification range, rich sample types, almost no damage and pollution to the original sample, and can obtain the morphology, structure, composition, and crystallographic information at the same time.	[35,65,66]
Spectroscopy technique	X-ray absorption spectroscopy (XANES)	It not only reflects the geometrical configuration of atoms in the surrounding environment of the absorbing atoms, but also reflects the structure of the low-energy electronic state near the Fermi level of condensed matter, so it becomes a useful tool for studying condensed matter.	[30,35,43]
	Extended X-ray absorption fine structure spectroscopy (EXAFS)	It is used to examine the local environments of central metal atoms, such as coordination numbers of atoms, distance between atoms, and adjacent atomic species.	[30,43]
	Infrared (IR) spectroscopy	It is widely used in the research of catalyst surface performance. The spectrum can provide information about the metal species by comparison with a reference compound. IR can also provide evidence of symmetry of certain samples, which can infer the bonding information between the metal and the support.	[67]
Other experimental characteristics	X-ray diffraction (XRD)	Different crystal composition determines the diffraction characteristics of the specific, including diffraction direction and intensity of diffraction, the diffraction spectra and reflection of the particular-crystal structure. It can also be used to determine the crystal cell parameters and space group, atomic specie and distribution, etc.	[30,64–66,68]
	BET	The BET specific surface area test can be used to measure the specific surface area, pore volume, pore diameter distribution and nitrogen adsorption and desorption curves of particles. The determination of BET is necessary for nanomaterials, especially for catalyst materials.

Experimental techniques Electron microscopy approach

High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) can be used to accurately observe the position of the metal atoms on the substrate, which is the most direct way to prove dimers or single clusters obtained.^[55] Besides spatial distribution, this technique is also used to estimate the crystallographic relationship regarding the surface structure of supports. For example, through the TEM image of the (Fe, Co)/NC in the light and dark areas and the corresponding size/shape evolution, Wu et al.^[53] directly traced that the metal-imidazolate-metal bond is continuously decomposed. As shown in Figure 4A,C, the resulting internal cavity and size increased due to the decomposition. Subsequently, the use of HAADF-STEM revealed that Fe and Co atoms existed in the porous carbon matrix as evenly distributed small bright double points (Figure 4B,D). Yan and co-workers^[48] used HAADF-STEM images to illustrate the morphology of the monoatomic Pt₁/graphene and diatomic Pt₂/graphene catalysts. After the first ALD cycle at 250℃ (Figure 4G), neither Pt clusters nor nanoparticles were observed. And the distance between Pt₁ single atoms was greater than 2 mm, which indicated that MeCpPtMe₃ molecules served as nucleation sites to form Pt₂ dimers in the second ALD cycle. According to HAADF-STEM images (Figure 4E,F), Pt₂ dimer is mainly formed along the direction of graphene without the presence of Pt cluster and NPs. In addition, it can be observed by STEM measurement that the Pt₂ dimer splits to form two isolated Pt₁ atoms by rotating at different angles, which proves the existence of the Pt₂ dimer rather than the overlap of two Pt single atoms. Wang et al.^[43] used aberration-corrected HAADF-STEM images to determine that Fe₂ clusters uniformly were dispersed on the mpg-C₃N₄. The evenly distributed small bright spots on the mpg-C₃N₄ substrate are confirmed as Fe atoms owing to the significant difference in Z contrast between Fe and N/C. Moreover, in the magnified Aberration-corrected HAADF-STEM image (Figure 4H,I), most of isolated Pt dimers appearing in the area marked by white circles, confirming the formation of diatomic Fe₂ clusters.

FIGURE 4. (A, C) TEM images of Fe/Co-N-C, (B, D) HAADF-STEM of Fe/Co-N-C (Copyright 2017, ACS Publications). AC HAADF-STEM images of monoatomic Pt1/graphene (G) and diatomic Pt2/graphene (E, F). Scale bars, 20 nm (E), and 1 nm (F, G). (Copyright 2017, Springer Nature.) H, Enlarged AC HAADF-STEM images of Fe2/mpg-C3N4. Scale bar, 1 nm. I, Fe–Fe distance in the intensity profiles. (Copyright 2018, Springer Nature.)

Spectroscopy technique

In order to understand the excellent catalytic performance of DACs/SCCs, it is not only necessary to intuitively characterize the degree of dispersion of metal atoms on the support, but also to understand the microstructure of DACs/SCCs, including coordination structure and electronic structure information. X-ray absorption spectroscopy (XAS) based on synchrotron radiation is divided into near-edge structure X-ray absorption near edge structure (XANES) and complementary extended X-ray absorption fine structure spectroscopy (EXAFS),^[56] which is a powerful tool to characterize the geometric structure of the active sites of DACs/SCCs. For example, Hu et al.^[50] detected changes in the structure of catalytic sites during the activation of Co-Fe-N-C by XAS. Figure 5A,B show that the spectral characteristics of Co-N-C and Co-Fe-N-C are significantly different from the spectral characteristics of Co foil. The oxidation state of Co ions in Co-N-C is close to +2, indicating that the energy of the main absorption edge is equivalent to the energy of CoO (2+) and slightly lower than the energy of Co₃O₄ (2+,3+). After activation, the energy at the absorption edge increase confirms the formation of Co (3+). The EXAFS spectra at the Co K-edge further demonstrated the structural evolution and formation of the Co-Fe diatomic catalyst.

FIGURE 5. A, XANES spectra of Co K-edge, B, Corresponding figure (A) K-edge energy of XANES spectrum. (Copyright 2019, ACS Publications.) C, Fourier transforms (FT) for Pt monomers, Pt–Ru diatomic pairs and Pt foil. (Copyright 2019, Nature.) D, Wavelet transforms contour plots of synthesized Co-N-C sample. (Copyright 2018, Elsevier)

In addition to XANES spectra, EXAFS spectroscopy is also used to examine the local environment of central metal atoms, such as coordination numbers of atoms, distance between atoms, and adjacent atomic species. Sun et.al.^[49] conducted the Fourier transform (FT) test of Pt EXAFS on Pt single atom, Pt-Ru dimer and Pt foil (Figure 5C). As for the tested Pt monoatomic sample, no peaks were observed at 2.7 or 3.3 Å, which indicates that the possibility of forming Pt-O bonds can be ruled out. Consistent with previous studies,^[57–59] and a peak appears at 2.6 Å due to the formation of Pt-Pt or Pt-Ru bonds. Besides an obvious Pt-N/Pt-C peak, the relatively weak peak at around 2.6 Å is further resolved. For the Pt foil, the first shell FT peak resolved at 2.6 Å should be correspondent to Pt-Pt bond. Since the dimer structure is derived from a single Pt atom, the peak at 2.6 Å is formed by the Pt-Ru bond.

Moreover, the pressure-dependent diffuse reflectance infrared Fourier transform (DRIFT) measurement and wavelet transform (WT) can further investigate the oxidation state and coordination structure of DACs and SCCs.^[60–62] WT-EXAFS is the perfect complement to FT. It is used to distinguish backscattered atoms and provide reliable information, such as the valence of the metal center. Even if the adjacent chemical bonds overlap, it can still provide powerful resolution in k-space and R-space.^[63] For instance, Xiao et al.^[26] studied a Co₂ diatomic site through WT-EXAFS. The WT contour plots of synthesized Co-N-C sample showed the coexistence of a Co–N path in the bottom contour map (A in Figure 5D) and a Co–Co path in the top contour map (B and C in Figure 5D). The WT-EXAFS revealed that the k values of B (6.88 Å^–1) was smaller than that of C (7.09 Å^–1), indicating that the Co-Co distance was shortened compared with cobalt nanoparticles. In addition, complementary comparison of the q space size of the k²-weighted EXAFS paths is calculated by FEFF, which indicated that the slight negative shift of the maximum WT of Co-NC sample is caused by the shortening of the Co–Co bond distance (2.12 Å).

Other experimental characteristics

X-ray diffraction (XRD) method is widely used in structural characterization of solid materials. Different crystal compositions determine the diffraction characteristics of the species. And their diffraction patterns have obvious differences in the number of diffraction peaks and the shape of the diffraction peaks. The atomically dispersed metal catalysts can be determined by confirming no X-ray diffraction spectrum of crystalline metals in the prepared samples.^[30,64] Moreover, the measurement of BET is also necessary for nanomaterials, especially for catalyst materials. The commonly used measurement method of specific surface area is nitrogen adsorption BET method.

DFT calculations

Density functional theory (DFT) method has been developed in parallel with experiments and it is currently indispensable for the studies of DACs and SCCs. First, DFT calculations could obtain charge distribution, adsorption energies and reaction mechanism, which can be used to explain the experimental phenomena. Second, DFT calculations can be used to predict and rationally design highly efficient DACs/SCCs.^[69–71]

DFT calculations have been commonly used to estimate the electronic properties of catalytic sites and explain the impact of some key structures on catalytic performance. Zhao et al.^[66] reported a diatomic metal-nitrogen site (Ni/Fe-N-C) via an ionic exchange strategy for efficient CO₂ reduction reaction (CO₂RR). With the assistance of DFT calculations, the free energy diagram for CO₂RR to CO and the CO₂RR mechanism of bimetallic nitrogen sites were established (Figure 6A). The Ni-Fe-NC structure with adsorbed CO was also calculated to provide an additional active site either on Fe or Ni for the second CO₂ activation. Compared with bare Ni/Fe-N-C, the bonding strength of COOH^* and CO^* on the Ni/Fe-NC structure with CO adsorbed is weak. It adds the free energy barrier of CO₂ (g) → COOH^* to 0.47 eV, whereas decreases the barrier of CO^*→CO(g) to 0.27 eV. Therefore, the theoretical overpotential of Ni/Fe-N-C for CO adsorption is reduced from 0.76 to 0.47 V, which greatly reduces the energy barrier for CO production, resulting in excellent CO₂RR activity. Zhou et al.^[32] successively calculated the adsorption strength of CO₂ and CO on the trimeric metal cluster M₃@NG by DFT, which determines possible reaction pathways and evaluating product selectivity (Figure 6B). The trimeric metal cluster M₃@NG composed of various elements can co-adsorb two CO₂ molecules with the adsorption energy of 1.22-6.66 eV, which were higher than the 3d transition metal dimer. Only Cr₂, Mn₂ and Fe₂ dimers had the ability to reduce two CO₂ molecules at the same time, with the adsorption energy of 0.48−0.76 eV. The adsorption energy of co-adsorbing two CO molecules on the M₃@NG system was 2.72-8.70 eV, which indicates CO strong adsorption on metal trimers. Therefore, production of CO during CO₂ reduction would be suppressed on these cluster catalysts. These clustered catalysts paved the way for further reduction and CC coupling reactions. Moreover, the metal trimer on the graphite carbon substrate can adsorb the third CO₂ molecule. The kinetic barriers of C₂C coupling energy of mostly M₃@NG were in the range of 0.10−0.65 eV, which would be beneficial to the selective production of C₃ hydrocarbons and alcohols.

FIGURE 6. A, The calculated free energy diagrams for CO2RR to CO on different catalysts. (Copyright 2019, Wiley-VCH.) B, Schematic illustration of the most efficient pathways for CO2RR toward C1-C3 products on M3@NG. (Copyright 2020, Elsevier.)

DFT methods can perform structural simulation and rationally design catalysts, which enable the rational construction of stable catalysts and better bridge the gap between theory and experiments. Huang et al.^[38] constructed the activity map of DACs through large-scale systematic DFT computations. Two-dimensional extended phthalocyanine (Pc) was used as a substrate to construct homonuclear (M₂-Pc) and heteronuclear (MM-Pc) DACs. The DAC activity map constructed using N₂H* adsorption energy as the activity descriptor can greatly decrease the number of candidate catalysts. They evaluated the thermodynamic and electrochemical stability of these homonuclear M₂-Pc. As can be seen in Figure 7A, 25 homonuclear M₂-Pc meeting the stability criteria were screened, which showed the reliability and feasibility of the experimental method. Then, they used the limiting potential (U_L) to evaluate the activity trend of 25 homonuclear M₂-Pc, setting the corresponding U_L on Ru (0001) (cat. - 0.98 V) as a metal-based benchmark. Compared with the stepped Ru (0001) surface, Ti₂-Pc, V₂-Pc and Re₂-Pc have lower U_L values (−0.75, −0.39 and −0.82 V, respectively), which display excellent efficiency to the nitrogen reduction reaction (NRR). When ΔE_NH2* was regarded as the descriptor to predict the catalytic behavior of homo-paired M₂-Pc, the V₂-Pc with the highest activity located near the volcanic peak. This was consistent with the results from DFT calculation mentioned above, which showed that ΔE_NH2* is a suitable activity descriptor to screen effective DACs for NRR. Finally, the hetero-paired MM'-Pc with better activity was also screened by using ΔE_NH2* as the activity description, and the activity diagrams of M₂-Pc and MM'-Pc were obtained (Figure 7B). Wang et al.^[72] studied 21 kinds of heteronuclear transition metal dimers as potential diatomic catalysts embedded in monolayer C₂N. With the assistance of DFT calculations, the selected CuCr/C₂N and CuMn/C₂N heteronuclear diatomic catalysts broke through the stubborn restriction of scaling relations on CO₂RR. By adjusting the composition of the dimer, two different types of metal atoms act as carbon adsorption site and oxygen adsorption site respectively due to the decoupling of key intermediates. Further analysis of the adsorption relationship for key intermediates, CO₂ can be effectively reduced to CH₄ on CuCr/C₂N and CuMn/C₂N. As shown in Figure 7C, the free energies of CuCr/C₂N and CuMn/C₂N were increased by 0.37 and 0.32 eV, which significantly break the *CO-*CHO scaling relationship. And the corresponding overpotential were only 0.54 V (CuCr/C₂N) and 0.49 V (CuMn/C₂N), which were greatly lower than the pure transition metal surface (for example, the theoretical overpotential of Cu surface is 0.91 V).

FIGURE 7. A, Stability test of M2-Pc, B, Variations of ΔEN2H* on heteronuclear MM'-Pc. (Copyright 2020, ACS Publications.) C, Free energy diagram of the CO2RR reduction pathway of CuCr/C2N at –0.37 V and CuMn/C2N at –0.32 V. (Copyright 2020, RSC)

THE CORRELATIONS BETWEEN THE STRUCTURE AND REACTIVITY OF DACS AND SCCS

When the catalyst size is reduced to the cluster and atomic-level distribution, leading to an increase in the unsaturated coordination environment of metal species. This unique structural feature makes DACs/SCCs often undergo fundamental changes in electronic structure, and accordingly show different catalytic performances from nano-catalysts. This chapter summarizes the unique structure of DAC/SCC from three aspects: support effect, coordination effect, and synergistic effect between metal atoms, and has an in-depth understanding of the relationship between DACs/SCCs’ structure and catalytic performance.

Support effect

The role of the support material in the catalytic process is not only to anchor the active species, but it can help to optimize the local geometry and electronic structure of the metal atoms.^[73–75] Therefore, choosing the appropriate support material plays a vital role in the excellent performance of DACs/SCCs.^[76] There are many reports the supporting material for DACs/SCCs, among which are N-doped carbon materials and metal organic frameworks (MOFs) very well-liked.

N-doped carbon materials

Previous researchers have found that N-doped carbon materials can activate the activity of the carbon skeleton, and effectively anchor metal atoms. Incorporating non-metal atoms in the matrix can effectively capture metal atoms to achieve high metal loading, because these non-metal atoms have lone pairs of electrons to provide additional coordination sites for metal atoms. Besides, different numbers of coordination atoms and coordination environments can induce changes in the electronic structure of the metal active sites, resulting in differences in activity and selectivity.^[77–79] C₂N, a novel nitrogenated holey graphene, with uniformly distributed macropores provides an opportunity to anchor more than one metal atoms. It has attracted much attention due to its continuous network structure with high dynamic stability and conductivity.^[80–82] As reported, Zhao et al.^[83] systematically studied the catalytic properties of several transition metal dimers (M₂@C₂N, M = Cu, Fe, etc.) supported on the C₂N layer as CO₂RR electrocatalysts. The 2p orbital of the N atom doped on the C₂N can be coupled with the 3d orbital of the metal atom Cu, preventing metal drift, thus Cu diatomic pairs can be stably anchored in the porous C₂N layer. Moreover, Cu has moderate adsorption energy for CO₂RR intermediates due to the coupling between N and Cu, making it have higher catalytic performance for reducing CO₂. Zhang et al.^[84] designed a series of C₂N layer anchoring transition metal atom catalysts (TM₂@C₂N, TM = Mo, Co, Ti, Mn, Fe, Cu) and explored their electrocatalytic performance for NRR. Among them, Mo₂@C₂N showed the best catalytic performance. In addition to NRR, TM₂-C₂N catalyst can also enhance the electrocatalytic efficiency of hydrogen evolution reaction (HER), oxygen evolution reaction (OER), and reducing oxygen reaction (ORR).^[85] Analogously, g-C₃N₄, a sheet-like network structure similar with graphene, is also used as a support material for diatomic catalysts. For example, Wang et al.^[43] synthesized highly dispersed Fe₂ clusters and chose to load these clusters on mesoporous carbon nitride (mpg-C₃N₄). Pyridine-N was evenly distributed on the top of the macropores of mpg-C₃N₄, which can provide many lone pair electrons to capture metal ions. Furthermore, Zheng et al.^[107] used a nitrogen coordination strategy to enable the atomic Cu species (Cu-N_x) on the nitrogen-doped carbon framework to be well dispersed and attached. The Cu doping concentrations and Cu-N_x configuration can be adjusted well by changing the calcination temperature. The Cu concentration can be as high as 4.9% _mol at a pyrolysis temperature of 800°C, two CO molecules were bounded on two Cu-N₂ positions since the distance between Cu atoms was close enough. This synergistic effect between them can be used to induce the C−C coupling and produce C₂H₄. When the pyrolysis temperature was higher than 800°C, more atomic Cu species and N species detached from the carbon frameworks. At the Cu concentration was lower than 2.4% _mol, the distance between isolated Cu-N₂ and the neighboring Cu-N₄ species was relatively large, suggesting that unable to favor the C−C coupling and facilitates the formation of CH₄ as a C₁ product.

Metal organic frameworks (MOFs)

Metal organic frameworks (MOFs) are a new type of three-dimensional ordered porous crystalline materials. Their high porosity is conducive to promoting the mass transfer process in the reaction process, thereby improving the catalytic efficiency.^[86–88] MOFs can easily and precisely control the structure of the immobilized diatomic and cluster metal catalytic sites and promote catalysis, which has proved to be an ideal substrate for DACs or SCCs.^[51] Prado and co-workers^[89] reported the preparation of Pd₄^0/1+ clusters confined in negatively charged MOF via an ion exchange strategy. First, the researchers synthesized a negatively charged MOF structure of the framework, expressed as Mg₂[Mg₄[Cu₂(Me₃mpba)₂]₃]∙45H₂O. Through a two-step process ion exchange, all Mg²⁺ in the MOF crystal can be exchanged for Ni²⁺ and [Pd(NH₃)₄]²⁺. Finally, the MOF materials containing Pd₄ clusters were obtained by the reduction of NaBH₄. The above structural transitions were confirmed by X-ray single crystal diffraction analysis as shown in Figure 8B,C. Interestingly, the four Pd atoms present a one-dimensional linear distribution in this material. As a result, the resulting Pd₄-MOF catalyst has excellent catalytic performance in the carbene-mediated diazoacetate reaction, with a high yield (>90%), and the highest turnover rate (up to 100,000). Moreover, the metal nodes atom-dispersed in MOFs are conducive to coordinate coordination environment,^[53] which makes them an ideal type of precursor for the construction of DACs/SCCs.^[54] Li et al.^[52] encapsulate and separate the preselected metal cluster precursors through molecular-scale zeolite imidazolium salt framework molecular-scale cages (ZIF), and then pyrolysis to remove the ligands of the Ru₃(CO)₁₂ precursors to produce Ru₃ decorated porous nitrogen-doped carbon cluster (Ru₃/CN). The experimental characterization shows that Ru-N and Ru-Ru coordination bonds are found to coexist, which means that Ru₃ clusters are uniformly fixed on the nitrogen-doped porous material. Importantly, the experimentally measured Ru₃ clusters showed conversion efficiency as high as 100% after 0.5 hours, and its TOF was 4320 h^–1 about Ru particles 23 times.

FIGURE 8. Schematic diagram of PtRuSA anchored in N-vacancy-rich g-C3N4. (Copyright 2019, ACS Publications)

Other supports

One-dimensional alloy nanowires have inherent structural advantages, such as high ratio branching, fast electron and substance transfer, low solubility and not agglomeration,^[90] which make them promising to be substrates for DACs or SCCs. Li et al.^[16] demonstrated that Cu atom pairs can be stably anchored on Pd₁₀Te₃ alloy nanowires. Besides, metal oxides are also commonly used as substrates for supporting atomic-level dispersion catalysts, because they can effectively inhibit the agglomeration of metal atoms by forming complexes on the surface.^[71] Zeolite is a nanoscale inorganic porous material. Its small pores can greatly accelerate the diffusion of guest molecules into and out of the material.^[68] It is usually employed as a substrate for atomic-level catalysts because of the strong acidity and high hydrothermal stability.^[91]

Coordination effect

Once the metal species is anchored to support, the strong metal-support interaction controls the electrocatalytic activity and selectivity for the target reaction. However, various coordination structures are formed during the preparation of DACs/SCCs,^[92] which will cause differences in metal-support interactions. Analyzing the coordination mode of metal atoms is a key step to establish a clear correlation between structure and activity.^[93,94] The coordination effect not only affects the structural and electronic properties of the metal centers, but also adjusts the catalytic reaction pathways.^[95–101] This section reviews the coordination effect from two aspects, named by the construction of surface defects and the modification of functional groups.

Surface defects

Compared with a perfect substrate, defective sites in support lead to the charge localization. The presence of these surface defects can help adjust the adsorption strength of key intermediates. Besides, these defects may change the surrounding electronic structure, and help to improve catalytic activity.^[41,102,103]

Guo and co-workers^[104] created a lot of N vacancies on the surface of C₃N₄ nanosheets via treatment at 620℃ in a reducing atmosphere, and made Pt-Ru dual sites embedded on N vacancy-rich g-C₃N₄ (PtRuSA-CN620) for the catalytic CO oxidation (Figure 8). Through STEM analysis, there were many adjacent Pt-Ru dimers on the PtRuSA-CN620, while Pt-Ru loaded on the perfect g-C₃N₄ surface (PtRuSA-CN) formed many PtRu alloy clusters. It implied that the existence of N vacancies was beneficial to the formation of the Pt-Ru diatomic center. And the results from DFT calculations confirmed that the C-Pt-Ru-N coordination structure was more stable due to Pt-C bonds having the lower adsorption energy. It indicated that the existence of nitrogen vacancies changed the PtRuSA-CN coordination environment N-Pt-Ru-N. Furthermore, the Pt atom in the C-Pt-Ru-N was an electron-rich center by analyzing XPS and XAFS, which was beneficial to catalyze some reduction-dependent reactions. Additionally, Li et al.^[105] reported a bimetallic SCC Rh₁Co₃/CoO(011) for the biomimetic N₂-NH₃ thermal conversion, in which the isolated Rh atom was doped at oxygen vacancies and coordinated with three Co atoms. Due to the existence of oxygen defects on the CoO surface, the Rh atoms in the Rh₁Co₃ bimetallic sites lay in a low oxidation state and provide a lot of electrons for adsorption molecules N₂, which further weaken the N≡N bond and promote the activation of N₂. Besides, the constructed Rh₁Co₃ structure drove a unique association reaction pathway, in which N₂ and H₂ could be co-adsorbed on the unsaturated Rh center and the two N atoms of N₂ are alternately hydrogenated at bimetallic active sites. As a result, the catalytic ability of the Rh₁Co₃ catalyst came from not only this electronic effect, but also the complementary role of the synergistic metal Rh in catalysis.

Modification of functional groups

Aside from the surface defects, the functional groups (e.g., OH and CO) coordinated with metals can also change electronic structure of DACs and SCCs, which affects the reaction behavior of other reactants on the surface and improves catalytic performance. Compared with large sized metal nanoparticles, the electronic structure of DACs and SCCs with highly unsaturated coordination is more susceptible to the influence of external functional groups.

Zheng et al.^[106] studied the adsorption of H on the Pd_n (n = 2-147) clusters and its hydrogenation performance through the DFT theoretical calculation. The strong adsorption of reactants on ultra-small Pd metal clusters (such as Pd₂/TiO₂ and Pd₃/TiO₂) inhibited catalytic hydrogenation, while the introduction of CO weakened the binding strength of hydrogen and hydrogenation intermediates, significantly improving hydrogenation activity. The reason was that the electronic structure of Pd₂/Pd₃ metal clusters can be rearranged by introducing CO molecules with strong electron-withdrawing ability, which may improve the catalytic hydrogenation performance of Pd clusters. Additionally, Xing et al.^[107] found that the introduction of an electron-withdrawing OH ligand at the center of the Fe-Co diatom acted as an energy level modifier, which can adjust the adsorption strength of the intermediate. And the triangular Fe-Co-OH coordination promoted the breaking of O-O bond, so that the catalytic activity of the constructed FeCoN₅-OH site was more than 20 times higher than that of FeN₄.

Metal atom synergistic effect

Unlike SACs, the biggest advantage of supported metal complex-site catalysts is the interaction between adjacent metal atoms.^[108] Interaction between metal atoms in DACs/SCCs helps to adjust the electronic structure of catalytic active sites, effectively improving the catalytic performance.^[64,109,110] This section reviews the synergy between metal atoms from two aspects: DACs and SCCs.

Synergistic catalysis in DACs

In DACs, two adjacent metal atoms are bonded to each other, and the electronic structure of the two metal sites can influence each other, which mainly promotes catalytic performance by adjusting the binding energy of reaction intermediates. DACs can be further classified into two subcategories: (1) homo-paired DACs with identical metal atoms, and (2) hetero-paired DACs with different metal atoms.

Synergistic catalysis by homo-paired DACs can be applied to various activation reactions. In CO₂-related activation reactions, Chen et al.^[16] designed the Cu₁⁰-Cu₁^x+ atom-pair was stably anchored on one-dimensional Pd₁₀Te₃ alloy nanowires. The EXAFS fitting results showed that the Cu-Cu coordination number was 2.1 and smaller than that of Cu foil, suggesting that Cu existed in a form of well-isolated small clusters Cu_x in the defects of Pd₁₀Te₃. In CO₂RR, Cu species in situ XANES spectra retained their stability under an electrochemical potential of −0.98 V. Compared with the traditional Cu catalyst and the Cu embedded in the Cu₂O catalyst, DFT calculations showed that the Cu₁⁰-Cu₁^x+ atom pair has the most favorable kinetics and thermodynamics and it can reduce the size of the catalytic interface to a metal diatomic pair. Cu₁⁰ can adsorb CO₂ molecules stably, while the adjacent Cu₁^x+ chemically adsorbed H₂O molecules. Therefore, the Cu₁⁰-Cu₁^x+ pair worked synergistically to promotes CO₂ activation. Besides, Zeng et al.^[34] prepared Pt/MoS₂ with adjacent Pt atomic structure to catalyze CO₂RR reaction, wherein Pt dimer maintained its atomic level dispersion under 7.5% high metal loading. Compared with low-load platinum single-atom catalysts, the synergistic effect of adjacent Pt atoms on MoS₂ reduces the activation energy of the reaction, leading to significant catalytic activity increase of CO₂ hydrogenation. More importantly, the synergistic action of two adjacent Pt atoms changes the pathway of the reaction. Only when Pt atoms were separated, CO₂ was directly converted to CH₃OH without HCOOH intermediate. In contrast, at the Pt diatomic site (Pt-S-Pt), CO₂ was converted into HCOOH in the first step. HCOOH was then further hydrogenated into CH₃OH and the path change greatly improved the CO₂ conversion rate. Thus, the synergetic interaction of the metal dimer in DACs may greatly improve catalytic performance and selectivity in CO₂ hydrogenation. Homo-paired DACs also can catalyze other reactions. Xiong et al.^[37] revealed that Fe-N-C catalysts had high catalytic activity and selectivity for ORR through high-temperature calcination. The formation of clusters with different numbers of iron atoms in ZIF-8 can accurately obtain Fe₁, Fe₂ and Fe₃ clusters, respectively. Among them, Fe₂-N-C catalyst showed the best catalytic activity, which is close to commercial Pt/C catalyst. The half-wave potential is 0.78 V and much higher than that of Fe₁-N-C and Fe₃-N-C catalysts (Figure 9A). Under acidic conditions, the half-wave potential of Fe₂-N-C only dropped by 20 mV after 20,000 cycles, which was an indicative of good cycle stability (Figure 10B). Further analysis of the reasons for the high catalytic activity of Fe₂ revealed that the adsorption mode of oxygen molecules on the surface played a direct role in the activity of oxygen reduction. As the number of Fe atoms in the cluster increasing, the distance between Fe atoms in Fe₂ and Fe₃ clusters was shortened relative to that of Fe₁ atom-dispersed. The theoretical calculation results (Figures 10C-E) indicated that O₂ molecules adsorption behavior in Fe₁ site followed superoxo-like model, while that in Fe₂ and Fe₃ sites was peroxide-type model. Compared with superoxo-like O₂, peroxo-like O₂ possesses higher adsorption energy, and O₂ was easier to be activated due to the extended O-O bond. To sum up, synergetic effect of diatomic sites improves the reaction activity and selectivity not only by changing the adsorption mode of the reactants, but also by adjusting the reaction path.

FIGURE 9. A, Fex-N-C (x = 0, 1, 2, 3) half-wave potential and starting potential comparison. B, ORR stability tests of Fe2-N-C. C, Superoxo-like adsorption at Fe1-N-C. D, Peroxo-like adsorption at Fe2-N-C. E, Peroxo-like adsorption at Fe3-N-C. (Copyright 2019, Elsevier)

FIGURE 10. A, Optimized structure of O2 adsorbed on the ZnN4, CoN4 and ZnCoN6(OH) systems. B, ORR Linear scan voltammogram (LSV) curves. C, Free energy diagram for ORR of ZnN4, CoN4 and ZnCoN6(OH) system under U = 0.40 V acidic conditions. D, Free energy diagram for ORR of ZnN4, CoN4 and ZnCoN6(OH) system under U = –0.83 V alkaline conditions. (Copyright 2019, Wiley-VCH)

The synergistic effect of hetero-paired DACs has also been identified during catalyzing various reactions. Chen et al.^[111] systemically investigated the catalytic properties of the heteronuclear DAC Fe₁Cu₁@C₂N and homonuclear DACs (Fe₂@C₂N and Cu₂@C₂N) for CO oxidation by DFT calculations. Compared with Fe₂@C₂N and Cu₂@C₂N homonuclear dimers, the Fe₁Cu₁@C₂N heteronuclear dimer possesses high stabilities and activities toward CO oxidation at room temperature without suffering from the CO-poisoning problem. The Fe-Co diatomic catalyst prepared by Wang et al.^[37] can catalyze the CO oxidation reaction at a low temperature of –73℃ and has a high conversion frequency (TOF) of 0.096 s^–1. The d states of Co and Fe atoms are mainly located near the Fermi level, indicating that the Fe-Co sites can act active centers for adsorption and reaction. The charge density distribution showed that Fe atoms have a higher charge density than Co atoms, indicating a reducing nature of the Fe atom and an oxidizing nature of the Co atom. DFT studies showed that CO preferentially adsorbs at the Co sites and O₂ adsorbs at the Fe sites following the Langmuir-Hinshelwood (L-H) mechanism. Sun et al.^[112] synthesized the Zn/Co-Nx-C atom-dispersed diatomic high activity oxygen reduction catalyst via a competitive coordination strategy. HAADF-STEM and XAFS characterization confirmed that the Zn-Co diatomic pair exists. DFT calculations showed that the electronic structure of the catalyst is adjusted due to the coordination of the bimetallic Zn and Co with N. Compared with ZnN₄ and CoN₄, the adsorption of O₂ on ZnCoN₆ (OH) can significantly increase the O-O length (from1.23 to 1.42 Å) (Figure 10A). This could facilitate the cleavage of O-O bond and reduce the dissociation energy barrier. Thus, the ORR activity was effectively improved. The overpotential of ZnCoN₆ (OH) under alkaline conditions is 0.335 V, which is better than ZnN₄ (0.436 V) and CoN₄ (0.391 V) (Figures 10C,D). Besides, the Zn-Co diatomic pair under acidic conditions (initial potential: 0.97 V, half-wave potential: 0.796 V) (Figure 10B) also showed very excellent activity and stability.

Synergistic catalysis in SCCs

Unlike monoatomic and diatomic catalysts, the interaction of atoms will be more complicated in SCCs, including the overlapping of orbitals and the electronic structure change.^[113,114] The unique performance characteristics of single cluster catalysts depend on the number of atoms, nuclear nature, and synergistic effect.^[115–117]

Li et al.^[33] proposed a new concept of “Single-Cluster Catalysis,” which first revealed the feasibility of ammonia synthesis on a surface-stable isolated metallic cluster catalyst. Unlike the direct opening of the N≡N triple bond under high temperature and pressure in the current industrial catalytic synthesis of ammonia, this study found that N₂ on the Fe₃/θ-Al₂O₃ (010) surface clusters can be directly hydrogenated to form NNH intermediates under mild conditions. This new reaction mechanism is similar with the biological nitrogen fixation process in nature. The density of states (DOS) of adsorbed N₂ on Fe₂/θ-Al₂O₃ was studied to clarify the association mechanism. Since Fe₃ spin polarization provides higher exchange stability for most α spin orbitals, the energy level of Fe₃ α spin d orbital is about 2.5 eV and lower than that of N₂ π* orbital, suggesting no obvious interaction. And *N₂ with unpaired electron is active for hydrogenation. Therefore, the activation of N₂ is attributed to the large spin polarization on the Fe₃ cluster. By constructing a homogeneous single-cluster active center on the surface of a heterogeneous catalyst, this new association mechanism breaks the traditional dissociation mechanism of industrial ammonia synthesis and provides a new catalytic method for ammonia synthesis. Xing et al.^[118] conducted a theoretical study on a triatomic metal surface single cluster catalyst (SCC) with M_xM'₃-x/GDY (M, M' = Ru, Os) firmly anchored on the GDY. The four investigated catalysts (Ru₃, Os₁Ru₂, GDY-supported Ru₁Os₂ and Os₃) all showed strong stability and activity for hydrogenate acetylene to ethylene. As seen from Figure 11, taking the reaction path of acetylene hydrogenation by Os₃/GDY as an example. The Os₃ cluster combines the two reactants and triggers the reaction. Acetylene combines with two Os^b atoms, and H₂ dissociates at the Os^t site to form two isolated H atoms. The C-C distance of 1.45 Å is between ethylene (1.34 Å) and ethane (1.51 Å), suggesting that acetylene has great potential to be activated on Os₃/GDY. The first two hydrogenation barriers for acetylene to C₂H₄* are very low, but further ethylene hydrogenation has a higher energy barrier, resulting in high selectivity for the semi-hydrogenation of acetylene to ethylene. In this new type of SCCs, two metal atoms M^b mainly activate hydrocarbons, and the third M^t atom decomposes hydrogen molecules, providing an opportunity to synergistically tune the catalytic behavior of the hydrogenation reactions. This synergistic effect results in metal atoms of isolated clusters leached from different chemical environments due to the size mismatch between the metal clusters and the carbon framework.

FIGURE 11. Reaction energy curve of acetylene and ethylene hydrogenation step of Os3/GDY. (Copyright 2019, ACS Publications)

BRIDGING REMARKS BETWEEN SACS AND DACS/SCCS

By comparing SACs and DACs/SCCs, there are some similarities and differences. For common aspects, supported SAC and DAC/SCC are atomic-level catalysts and their catalytic efficiencies are mostly related to the structure of active center composed of metal atoms and coordination atoms. Both SACs and DACs/SCCs are desired to achieve high metal loading without aggregation, which depend on the choice of suitable substrate and preparation method.^[119] For different aspects, DACs/SCCs are originally derived from the design concept of single atom catalyst, while they have the following different aspects. First, the active sites of DACs/SCCs are extremely ultra-small metallic clusters, resulting in a higher metal loading than that of SACs. Second, ultra-small metallic clusters have a discontinuous band structure, where the band structure is closely related to the type and number of metal atoms. Changing the number of metal atoms in a small cluster or incorporating another metal element may greatly change its electronic structure. Therefore, the electronic structure of the active site of DACs/SCCs has more room for adjustment. Third, DACs/SCCs provide adjacent unsaturated coordination sites for multiple reactants adsorption and complicated reaction process. And the formation of unsaturated sites offers opportunities to lower the reaction energy barrier by changing ligand environments around metal center.^[120] Finally, apart from the interaction between metal atoms and supports, there is also a synergistic effect between metal atoms. The synergistic effect between metal atoms in DACs/SCCs can change the adsorption mode of the reactants and optimize the reaction pathway, thereby it can significantly improve the catalytic activity and selectivity.^[121] Thanks to the different characteristics listed above, compared with SACs, DACs/SCCs show good stability and high activity in various applications (such as hydrogenation and oxidation reactions) as shown in Table 2.

TABLE 2 The summary of the applications and performance comparison involved in DACs/SCCs and SACs

Application	Materials	Ref.
ORR	FeCoNX/C versus FeNx/C or CoNx/C FeCoNX/C exhibits 20 times the ORR activity of the mononuclear FeN4/CoN4 site, and the onset potential (Eonset) and half-wave potential (E1/2) are 1.02 and 0.86 V Fe2N6 versus FeN4 The planar-like Fe2N6 structure exhibits over 700% increase in mass activity than traditional isolated FeN4 sites.	[122] [53]
OER	Fe–NiNC-50 versus Fe-NC/ Ni-NC Fe–NiNC-50 exhibits higher OER activity than the Fe-NC and Ni-NC sites, with an onset potential (Eonset) and overpotential of 1.50 V and ca. 340 mV, which is lower than Ni-NC and Fe-NC, respectively. Co-Fe-N-C versus Co-N-C Fe is essential for the enhanced activity of Co-Fe-N-C compared with Co-N-C, and the activity of the former scales with the amount of Fe.	[44] [50]
CO oxidation	Ru–Pt monomers versus isolated Ru/Pt atoms The DFT results confirm that an electron-rich C-Pt-Ru-N structure is an optimized site for the catalytic CO oxidation compared with isolated Ru/Pt atoms.	[104]
CO2RR	Cu–atom-pair/Pd10Te3 versus isolated Cu atom Compared with Cu single atoms, H2O and CO2 molecules can be synergistically adsorbed on the Cu atom pair, which thereby promotes CO2 activation.	[16]
Alcohol oxidation	Ru3/CN versus Ru1/NC Higher conversion of 2-aminobenzyl alcohol for Ru3/CN (100%) compared with Ru1/CN (21%) are obtained. The TOFs of Ru3/CN (4320 h–1) is remarkably higher than that of Ru1/CN (416 h−1) at 20% conversion.	[52]
Hydrogenation	Ru1@ZIF-8 versus Ru3@ZIF-8 Higher conversions of phenylacetylene for Ru3@ZIF-8 (47%) compared with Ru1@ZIF-8 (18%) are obtained. MxM′3-x clusters (Ru3, Os1Ru2 and Os3) In SCC, two of the metal atoms mainly activate the hydrocarbon and the third one dissociates the H2 molecule. This synergistic effect is beneficial to improve the efficiency of the hydrogenation reaction.	[29] [107]

SUMMARY AND PERSPECTIVE

Atomic-scale catalysts have become the research frontier in the field of catalysis. The emergence of SACs has injected new vitality into this field with low cost and outstanding catalytic activity. However, SACs suffer from some limitations such as isolated single active site and low metal loading, which hindered the development of SAC. Derived from SACs, DACs and SCCs can add the number of active centers and promote its interaction with the supporting material, ensuring the dispersion of atoms. Moreover, the coordination between metal atoms could more effectively adjust the electronic structure of the d-orbital, and optimize catalytic performance. In this review, we summarized recent research progresses on DACs/SCCs, including preparation strategies, experimental characterization techniques and DFT methods to determine the active center and clarify the reaction mechanism. The effects of the microstructure of DACs/SCCs on the catalytic performance of DACs/SCCs are further investigated from three aspects, including carrier effect, synergy effect and coordination effect. At last, our systematic summaries of the similarities and differences between SACs and DACs/SCCs point out the superiority of DACs/SCCs to SACs.

Although some success stories have been achieved, many challenges are still faced in the research and practical applications of DACs/SCCs. First, some reports^[123,124] have indicated that a single-atom catalyst with high metal loading can be successfully prepared through constructing high density anchor sites. These synthesis methods are worthy of using for reference in the development of diatom/single cluster catalysts with high metal loading. In addition, the current evaluation criteria in experiment (hours or days) are far from meeting the requirements of practical applications on an industrial scale (in months or even years). This requires us to further screen out suitable combination of metals and supports, and improve the interactions between them.

Second, it is difficult to achieve precise control and synthesis of DACs and SCCs at the atomic level at an industrial scale. The synthesis of DACs/SCCs needs further research and optimization to obtain uniform active sites. It is also necessary to develop simple and convenient methods for large-scale synthesis of DACs/SCCs. For example, Li et al.^[125] reported direct bulk metals conversion into single-atom catalysts via gas migration strategy. Moreover, Ji et al.^[126] reported a simple and scalable ball milling method for the synthesis of kilogram-level coal-made Pt (Pt₁/Co) monoatomic alloy catalyst. Ma et al.^[127] has proved that the precious metal SAC can be mass-produced through a mechanochemical method. The research direction of DACs/SCCs can continue to develop toward the existing results of these SACs.

Third, how to accurately identify the evolution of structure and activity in DACs/SCCs is still a problem that needs further investigation. More precise operation and characterization technologies are needed to monitor the dynamic behavior of the electronic environment in real-time and determine the geometric structure of the active site, and the trajectory of the reaction intermediates.

Finally, the present development of DACs/SCCs is still based on the trial-and-error process. Although traditional DFT calculations are widely used, high-throughput screening is not suitable. The machine learning method that relies on data analysis of DFT calculations is expected to build predictive models, which demonstrates a deeper understanding of the structure-performance relationship and a direct high-throughput way catalytic performance prediction. And this would be of great help for the rational design of DACs/SCCs.

ACKNOWLEDGMENT

This work is supported by the China Postdoctoral Science Foundation (2019TQ0021).