Electrochemical CO₂ reduction catalyzed by atomically precise alkynyl-protected Au₇Ag₈, Ag₉Cu₆, and Au₂Ag₈Cu₅ nanoclusters: probing the effect of multi-metal core on selectivity

Abstract

Doping metal nanoclusters (NCs) with another metal usually leads to superior catalytic performance toward CO₂ reduction reaction (CO₂RR), yet elucidating the metal core effect is still challenging. Herein, we report the systematic study of atomically precise alkynyl-protected Au₇Ag₈, Ag₉Cu₆, and Au₂Ag₈Cu₅ NCs toward CO₂RR. Au₂Ag₈Cu₅ prepared by a site-specific metal exchange approach from Ag₉Cu₆ is the first case of trimetallic superatom with full-alkynyl protection. The three M₁₅ clusters exhibited drastically different CO₂RR performance. Specifically, Au₇Ag₈ demonstrated high selectivity for CO formation in a wide voltage range (98.1% faradaic efficiency, FE, at −0.49 V and 89.0% FE at −1.20 V vs. RHE), while formation of formate becomes significant for Ag₉Cu₆ and Au₂Ag₈Cu₅ at more negative potentials. DFT calculations demonstrated that the exposed, undercoordinated metal atoms are the active sites and the hydride transfer as well as HCOO* stabilization on the Cu–Ag site plays a critical role in the formate formation. Our work shows that, tuning the metal centers of the ultrasmall metal NCs via metal exchange is very useful to probe the structure–selectivity relationships for CO₂RR.

---

Introduction

The electrochemical CO₂ reduction reaction (CO₂RR) has been attracting increasing research efforts continuously, as it can convert CO₂ into valuable fuels and balance the carbon cycle.^1–4 So far, various metals including Au, Ag, Cu, etc. as catalytic materials have been investigated for CO₂RR.^5,6 Bimetallic or trimetallic catalysts usually exhibit superior catalytic performance than their homometallic counterparts due to the catalytic synergistic effects.^7,8 To improve the catalytic efficiency and advance the fundamental mechanistic understanding, one of the major challenges is the polydispersity of the catalyst. Specifically, despite the size, morphology, composition, even the coordination environment seems to be uniform in bulk or at a large-scale dimension, it can't offer a homogeneous chemical environment at the atomic level, making it extremely challenging to profoundly elucidate the mechanism and establish the structure–function relationship.

The emergence of atomically precise coinage metal nanoclusters (NCs) offers great opportunities to resolve the above problem due to their definitive size, morphology, composition, and more importantly, the crystallographically resolved structure can provide well-defined chemical environment to correlate the structure–performance relationship.^9–16 Pioneering work has been extensively conducted on thiolate-protected bimetallic NCs. For instance, in an early study, Jin group discovered that, compared to homogold Au₂₅ NC, monopalladium-doped Pd₁Au₂₄ NC can drastically inhibit the H₂ evolution, and had much higher CO product selectivity (faradaic efficiency for CO, FE_CO = ∼100%) at high potentials.¹⁷ Zhuang et al. found that, compared with the parent Au₄₄ NC, Au₄₇Cd₂(TBBT)₃₁ (TBBT: 4-tert-butylbenzenelthiol) NC exhibited not only higher selectivity for CO (FE_CO up to 96% at −0.57 V), but also a higher CO partial current density (j_CO = −3.67 mA cm⁻²) with a stronger suppression of the hydrogen evolution reaction (HER) (FE_H2 = ∼3.8%).¹⁸ In another study, by only substituting four surface Au atoms in Au₂₃(SR)₁₆ with two Cd atoms, Au₁₉Cd₂(SR)₁₆ was prepared by Li et al., and such modification greatly enhanced the selectivity of CO in CO₂RR (FE_CO = ∼90 to 95% at −0.5 to −0.9 V), which is doubled compared to the undoped Au₂₃ NC.¹⁹ Recently, Sun et al. devised a strategy to control the cleavage of Au–S or S–C bonds by introducing Cd atoms, and identified the reaction sites of Au₂₅(SR)₁₈, Au₂₄Cd₁(SR)₁₈, Au₁₉Cd₄(SR)₁₈, and Au₃₈Cd₄(SR)₃₀ for CO₂RR.²⁰ In the above cases, DFT calculations disclosed that, the Cd doping altered the surface geometry and electronic structure of the NCs, which further changed the intermediate binding energy.

Noteworthily, for Au-based NCs, CO is the main product in CO₂RR test. Copper-based catalysts have demonstrated to be effective to convert CO₂ into highly valuable products including formate,²¹ methanol,²² methane,²³ and so on. Tang et al. synthesized a Cu₃₂H₂₀L₁₂ (L: a dithiophosphate ligand) NC, which can offer a unique selectivity of formate (FE_formate = 90%) for CO₂RR at low overpotentials.²⁴ DFT calculations revealed that, the presence of the negatively charged hydrides in the NC played a critical role in determining the selectivity of the product, while the formate formation proceeded via the lattice-hydride mechanism.²⁴ Thanks to the versatile metal–ligand bonding moieties,^25–27 alkynyl ligands have been attracting more and more attentions to prepare coinage metal NCs in the past decade,^25,28,29 and homoleptic alkynyl-protected coinage metal NCs possess unique physicochemical properties and have found broad applications in semiconductor,³⁰ hypergolic fuels,³¹ and biomedical regime.³² Until so far, significant progress has been made on structure determination and formation mechanism study,^25,28 yet the cases on alkynyl-protected metal NCs for CO₂RR are still quite rare. Recently, our group reported the quite small all-alkynyl-protected [Ag₁₅(C≡C-^tBu)₁₂]⁺ NC, which was able to convert CO₂ into CO with a FE_CO of ∼95% at −0.6 V.³³ Also, the first case on homoleptic alkynyl-protected AgCu superatom of [Ag₉Cu₆(C≡C-^tBu)₁₂]⁺ was prepared to compare the physicochemical properties with [Au₇Ag₈(C≡C-^tBu)₁₂]⁺, and the two M₁₅ clusters exhibited distinctly different optical properties due to the metal core difference.³⁴ The following questions arise immediately: will these two clusters have different CO₂RR performance as well? Furthermore, as both clusters are belonging to the M₁₅ series, if the metal core is atomically tailored, how does the CO₂RR performance change? In another word, can we atomically tailor the core to probe the metal core effect of the M₁₅ series toward CO₂RR? The above questions form the primary aim and goal of our current study.

Herein, we report the CO₂RR performance and comprehensive mechanistic study of atomically precise alkynyl-protected [Au₇Ag₈(C≡C-^tBu)₁₂]SbF₆ (Au₇Ag₈ in short hereafter), [Ag₉Cu₆(C≡C-^tBu)₁₂]SbF₆ (Ag₉Cu₆ in short hereafter), and [Au₂Ag₈Cu₅(C≡C-^tBu)₁₂]SbF₆ (Au₂Ag₈Cu₅ in short hereafter) NCs. As a note, in a recent study, Kang et al. reported a shortening of the A3-coupling reaction time from hours to minutes at higher temperatures (175 °C) catalyzed by a thermally robust, trimetallic Au₁Ag₁₆Cu₁₂(SSR)₁₂(PPh₃)₄ NC (SSR: benzene-1,3-dithiolate), demonstrating the unique potential of trimetallic alloying in catalytic enhancement.³⁵ By using a chiral reducing agent, Hakkinen and Zheng groups reported a novel phosphine and thiolate ligand co-protected trimetallic [Au₇Ag₆Cu₂(R- or S-BINAP)₃(SCH₂Ph)₆]SbF₆ (BINAP: 2,2′-bis(diphenylphosphino)-1,1′-binaphthyl) NC with tertiary chiral nanostructure.³⁶ In this study, Au₂Ag₈Cu₅ is the first case of all-alkynyl-protected trimetallic superatom documented so far. It can be synthesized by a metal exchange approach from Ag₉Cu₆, and X-ray crystallography reveals a body-centered-cubic (BCC) structure with an Au@AuAg₄Cu₃@Ag₄Cu₂ core configuration. Interestingly, the three M₁₅⁺ NCs exhibited significantly different CO₂RR properties. Au₇Ag₈ can convert CO₂ into CO exclusively with FE_CO reaching 98.1% at −0.49 V, while CO and formate are the main products for Ag₉Cu₆ and Au₂Ag₈Cu₅ at more negative potentials, in which the highest FE_formate value is 47.0% at −1.19 V and 28.3% at −0.99 V, respectively. In addition, Ag₉Cu₆ and Au₇Ag₈ can inhibit H₂ evolution effectively with FE_H2 less than 10% in the whole tested potential range. Density functional theory (DFT) calculations disclosed that –C≡CR removal from the intact NC can expose the undercoordinated metal atom as the catalytic site to significantly promote the activity and selectivity of CO₂RR. In particular, the formation of negative hydride is the key for the exclusive formate formation on Ag₉Cu₆ and Au₂Ag₈Cu₅.

Results and discussion

Preparation and characterization of Au₇Ag₈, Ag₉Cu₆, and Au₂Ag₈Cu₅ NCs

Au₇Ag₈ and Ag₉Cu₆ NCs were first synthesized by following the method in our previous report, in which the crystal structure and optical properties of the two NCs were compared.³⁴ Note that, the fabrication and total structure of Au₇Ag₈ NC was first reported by Wang et al. in 2016,³⁷ and our anti-galvanic synthetic approach can improve the yield drastically.³⁴ In this study, Au₂Ag₈Cu₅ NC was synthesized by a site-specific metal exchange method by a reaction between Me₂SAu(I)Cl and Ag₉Cu₆ NC with a controlled stoichoiometric ratio. The detailed synthetic procedure can be found in ESI,† and the relevant elucidation of the process will be discussed next.

Subsequently, the chemical composition of the three NCs were verified by electrospray ionization mass spectrometry (ESI-MS). As illustrated in Fig. 1a, the sharp peak at m/z = 3214.8510 and 2325.5677 is assigned to Au₇Ag₈ and Ag₉Cu₆, respectively, and the well-matched experimental/simulated isotopic pattern (Fig. S1a and b†) confirmed the molecular composition of Au₇Ag₈ and Ag₉Cu₆. In addition, the main peak at m/z = 2548.6640 corresponds well with [Au₂Ag₈Cu₅(C₆H₉)₁₂]⁺ (cal.: 2548.6634 Da, deviation: 0.0006 Da), and the isotopic patterns of the NC match perfectly with the simulated results (Fig. S1c†). There is also the one Au atom exchanged product of AuAg₈Cu₆ NC (indicated by ▲, see enlarged spectra in Fig. S1f†), and the fragments of Au₇Ag₈ and Ag₉Cu₆ NCs are identified in Fig. S1d and e,† respectively. To further confirm the metallic ratio in Au₂Ag₈Cu₅, X-ray photoelectron spectroscopy (XPS, Fig. S2†) and energy-dispersive X-ray spectroscopy (EDS, Fig. S3†) were conducted. The atomic ratio of Au/Ag/Cu is 1.9/8.1/5.3 (2.0/8.7/5.7) and 1.5/6.7/4.2 (2.0/8.9/5.5) from XPS (Table S1†) and EDS (Fig. S3†), respectively, both are in agreement with the theoretical value (2/8/5). The XPS survey scan spectra confirmed the presence of the essential elements (Fig. S2a†). The binding energy of the Au 4f_7/2 electrons is located at 84.43 eV, between bulk Au (84.0 eV)³⁸ and Au(I) (84.5 to 86.0 eV)³⁸ (Fig. S2b†). Furthermore, the binding energy of the Ag 3d_5/2 electrons is located at 368.65 eV (Fig. S2c†), indicating that the valence state of Ag atoms in Au₂Ag₈Cu₅ is +1.³⁹ In addition, the binding energy of Cu 2p_3/2 (933.40 eV) agrees well with that of Cu(I) (933.3 eV),⁴⁰ implying that Cu atoms are present as Cu(I) (Fig. S2d†). Moreover, as illustrated in Fig. 1b, the fingerprint absorbance peaks of Au₂Ag₈Cu₅ NC are located at 335, 461, 484, 521, and 571 nm, quite different from that of Au₇Ag₈ (313, 339, 422, 477, and 506 nm) and Ag₉Cu₆ (333, 357, 422, 544, 579, and 620 nm) NCs. Nevertheless, we monitored the absorbance change in the formation process of Au₂Ag₈Cu₅. As shown in Fig. S4a,† the absorbance feature of Ag₉Cu₆ NC disappeared immediately upon the addition of Me₂SAu(I)Cl, while a new absorption band at ∼484 nm arose. There is an obvious colour change at the timing point of Me₂SAu(I)Cl addition (Fig. S4b†). In 1 h, the characteristic peak at 484 nm from Au₂Ag₈Cu₅ gradually arose, meanwhile the absorbance peak at 571 nm can be identified. The metal exchange process occurs very fast, and as manifested by the two visualized video records (see Video 1† under room light and Video 2† under 365 nm UV-light as additional ESI†). In addition, we also studied the photoluminescence property of the Au₂Ag₈Cu₅ NC. As shown in Fig. S5,† Au₂Ag₈Cu₅ NC strongly emits in the near-infrared region (λ_max = 825 nm), which is much stronger than that of Au₇Ag₈ NC, while Ag₉Cu₆ NC is not photoluminescent.³⁴ Given the standard absorbance curve (Fig. S6a†) of Au₂Ag₈Cu₅ NC, according to Lambert–Beer's law, the molecular absorptivity (ε) of Au₂Ag₈Cu₅ NC can be determined (ε = 1.88 × 10⁴ M⁻¹ cm⁻¹), as summarized in Table S2.† Subsequently, the yield of Au₂Ag₈Cu₅ NC was calculated as ∼66.85% (based on Cu). The details of the calculation process can be found in ESI† (Fig. S6b and Table S3†).

Fig. 1 (a) Positive-mode ESI-MS and (b) absorbance spectra of Au₇Ag₈, Ag₉Cu₆, and Au₂Ag₈Cu₅ NCs. The asterisk (*) and octothorpe (#) indicate the fragment ion and molecular ion of Au₇Ag₈, Ag₉Cu₆, and Au₂Ag₈Cu₅ NCs, respectively, and the triangle (▲) indicates the AuAg₈Cu₆ NC product in the Au₂Ag₈Cu₅ sample.

Elucidating metal exchange process from Ag₉Cu₆ to Au₂Ag₈Cu₅

Tailoring the metal core while retaining the other parts has been proven as an effective approach to modify the physicochemical property and enhance the functionality of thiolate-protected metal NCs.⁴¹ Such metal core tailoring is actually part of the tailoring chemistry of metal nanoclusters.⁴² To the best of our knowledge, no case on metal core tailoring has been reported for homoleptic alkynyl-protected coinage metal NCs. Inspired by the findings in thiolate-protected metal NCs, herein, Au₂Ag₈Cu₅ was synthesized by a controlled stoichiometry of Me₂SAu(I)Cl-to-Ag₉Cu₆ (=2) through atomic-level tailoring by metal exchange (Fig. 2a). The total structure comparison between Au₂Ag₈Cu_5, Ag₉Cu₆, and Au₇Ag₈ will be discussed in the next session, nevertheless, the detailed five-step transformation from Ag₉Cu₆ to Au₂Ag₈Cu₅ is presented in Fig. 2b, and the corresponding chemical reaction equations are shown in Fig. 2c.

Fig. 2 (a) Synthesis of the [Au₂Ag₈Cu₅(C≡C-^tBu)₁₂]⁺ NC by atomic-level tailoring. (b) Metal exchange process from Ag₉Cu₆ NC to Au₂Ag₈Cu₅ NC via a five-step process. Color legend: Au, yellow; Ag, cyan; Cu, orange; C, gray, H, white. (c) Equations of the Au₂Ag₈Cu₅ formation by metal exchange from Ag₉Cu₆ NC. The atoms before (blue) and after (red) the reaction.

Firstly, 1 eq. of Me₂SAu(I)Cl was added to react with Ag₉Cu₆, and one Au(I) atom replaces one Ag(I) atom to form a Ag@Ag₇Au@Cu₆ kernel (Step I). It is a metathesis reaction, and Me₂SAg(I)Cl is also generated in the solution. Note that, the driving force of such heteroatom exchange is probably the interaction between the Cl⁻ ion and the Ag(I) atoms on the Ag₈ cube. Subsequently, the as-formed intermediate was transformed into more stable molecule (kernel: Au@Ag₈@Cu₆) via galvanic reaction, in which the Au(I) atom is reduced by the central Ag(0) atom (determined by DFT structures with Mulliken charges in Table S4†),³⁴ and the two atoms exchanged the position with each other (Step II). As a note, such Au heteroatom diffusion phenomenon has been previously documented in thiolate-protected alloy NCs,^43,44 for instance, Xie and coworkers discovered that, the Au heteroatom diffuses into the surface layer of the Ag₁₃ icosahedron kernel and finally is reduced by the central Ag(0) atom, forming thermodynamically stable AuAg₂₄(MHA)₁₈ molecule.⁴⁵ Then, in the presence of another 1 eq. of Me₂SAu(I)Cl, like the first step, Au₂Ag₇Cu₆ (kernel: Au@Ag₇Au@Cu₆) NC was formed by the metal exchange reaction (Step III). Consequently, the Cu atoms around the Au atom in the M₈ cube are activated and can react with Me₂SAg(I)Cl generated in the previous steps, and one Ag(I) atom exchanges with one Cu(I) atom in the Cu₆ octahedron (Step IV). Finally, three Ag(I) atoms on the M₈ cube were exchanged by three Cu(I) atoms to form Au₂Ag₈Cu₅ NC with the optimal thermodynamic stability (Step V).

It is worth pointing out that, the precise stoichiometric ratio of Me₂SAu(I)Cl-to-Ag₉Cu₆ (=2) is critical for yielding the optimal amount of Au₂Ag₈Cu₅ NC. In fact, different ratios of Me₂SAu(I)Cl (0.4 eq., 1.0 eq., 1.6 eq., 2.0 eq., and 2.4 eq. per Ag₉Cu₆) were tested, and the results are shown in Fig. S7.† As depicted in the absorbance change in Fig. S7a,† with the increasing of the Me₂SAu(I)Cl amount (from 0.4 to 2.0 eq.), the intensity of the characterstic peak (at 579 nm) from Ag₉Cu₆ decreased gradually (totally disappeared with 2 eq.), while the characteristic peak (at 484 nm) from Au₂Ag₈Cu₅ gradually became intensified. However, when it increased to 2.4 eq., the intensity of the peak at 484 nm slightly decreased. Such trend can be more clearly observed in Fig. S7b and c,† that is, with lower amounts of Me₂SAu(I)Cl (0.4 to 1.6 eq.), Ag₉Cu₆ NC is not fully converted, and with extra amount of Me₂SAu(I)Cl, Ag₉Cu₆ can be fully converted but polydispersed mixture was obtained. For instance, in the presence of 8 eq. Me₂SAuCl, a series of [Au_xAg₈Cu_7−x (C≡C-^tBu)₁₂]⁺ (x = 1 to 7) molecules including Au₂Ag₈Cu₅ and Au₇Ag₈ NCs were acquired, as confirmed by the ESI-MS spectra in Fig. S8.† Unfortunately, several attempts were conducted to separate the intermediate but was not successful, mainly due to that, this reaction occurs too fast (the whole process is finished in 1 h). We also noticed that, in the previous report, Wang et al. employed Cu atoms to react with Au₇Ag₈ clusters, and a series of cluster mixture [Cu_nAg₈Au_7−n (C≡C-^tBu)₁₂]⁺ (n = 0 to 6) including Au₂Ag₈Cu₅ NC (n = 5) was identified by mass spectrometry but the separation was also not performed neither.³⁷ Therefore, the exact stoichiometric ratio of 2 is the optimal value and also very critical.

Structural comparison of the three M₁₅ NCs

Subsequently, the atomic packing structure of Au₂Ag₈Cu₅ was examined by single crystal X-ray diffractometer (SC-XRD). As illustrated in Fig. S9,† Au₂Ag₈Cu₅ crystallizes in space group of R_3̄, and each unit cell has a SbF₆⁻ counterion, indicating that Au₂Ag₈Cu₅ NC possesses a +1 charge. The detailed structural parameters are summarized in Table S5.† The overall structure of monocationic Au₂Ag₈Cu₅ is shown in Fig. 3a, which contains two Au atoms, eight Ag atoms, five Cu atoms, and twelve tert-butylacetylene ligands, hence the molecule can be formulated as [Au₂Ag₈Cu₅(C≡C-^tBu)₁₂]SbF₆. As illustrated in the space-filling structure, one Au site, five Cu sites, and eight Ag sites on the surface of Au₂Ag₈Cu₅ are exposed partially, which might result in the differences in catalytic performance compared with the other M₁₅ NCs (Ag₉Cu₆ and Au₇Ag₈). As a note, all the currently reported alkynyl-protected M₁₅ NCs such as Ag₉Cu₆, Au₇Ag₈, and Ag₁₅ have only one type of ^tBu-C≡C-M-C≡C-^tBu (M = Cu/Au/Ag) linear motif.^33,34 However, for Au₂Ag₈Cu₅, there are five types of ^tBu-C≡C-M-C≡C-^tBu (M = Cu/Ag) motifs on the surface, in which the coordination mode of ^tBu-C≡C– ligands are μ₂-η₁ (Ag/Cu), η₁ (Ag1/Ag2) for motif 1; μ₂-η₁ (Au/Ag/Cu), η₁ (Ag3) for motif 2; μ₂-η₁ (Au/Ag/Cu), η₁ (Ag4) for motif 3; μ₂-η₁ (Au/Ag/Cu), η₁ (Cu1) for motif 4 and μ₂-η₁ (Au/Ag), η₁ (Cu2) for motif 5, respectively (Fig. 3b). As a result, the σ (Cu2–C) and π (Ag/Cu–C) bond lengths of motif 5 (average value: 1.870 Å and 2.389 Å) are shorter than those of motifs 1, 2, 3, and 4 (average value: 1.894 Å and 2.392 Å; 1.893 Å and 2.395 Å; 1.894 Å and 2.398 Å; and 1.873 Å and 2.394 Å, Table S6†). Motif 5 in Au₂Ag₈Cu₅ is nearly identical with the motif in Ag₉Cu₆. The average σ and π bond lengths of motifs on the surface of Au₂Ag₈Cu₅ (average 1.886 Å and 2.394 Å) are longer than those of Ag₉Cu₆ (average 1.870 Å and 2.381 Å) and shorter than those of Au₇Ag₈ (average 1.982 Å and 2.513 Å) (Table S7†). Note that, such different types of motifs can lead to the distortion of the kernel structure of Au₂Ag₈Cu₅. Furthermore, Au₂Ag₈Cu₅ adopts structural feature from both Au₇Ag₈ and Ag₉Cu₆, as it has the same central Au atom with Au₇Ag₈ and a more similar outlayer (Ag₄Cu₂vs. Cu₆) with Ag₉Cu₆ (Fig. 3c). Specifically, the anatomical structure of Au₂Ag₈Cu₅ is compared with the two bimetallic NCs. As shown in Fig. 3d, Au₂Ag₈Cu₅ adopts a core–shell–shell configuration (M_core@M_cube@M_octahedron) of Au@AuAg₄Cu₃@Ag₄Cu₂, similar to the other two NCs, but there are some difference in the M_cube and M_octahedron layers. For Au₂Ag₈Cu₅, the two layers consist of the heteroatoms (AuAg₄Cu₃ and Cu₂Ag₄), while there are the homoatoms in the middle layer (Ag₈) and outer layer (Au₆ and Cu₆) of Au₇Ag₈ and Ag₉Cu₆. Such structural difference leads to the difference in average bond lengths spread on different layers of the three NCs (Table S8†). Compared with Au₇Ag₈ and Ag₉Cu₆, the doped Au and Cu heteroatoms lead to the M_cube in Au₂Ag₈Cu₅ slightly relaxed, as the average adjacent M_cube–M_cube bond (3.414 Å) length is larger than those in the other two NCs (3.333 Å and 3.283 Å). However, there are four Ag atoms in the M_octahedron of Au₂Ag₈Cu₅, and the bonding lengths of the outer layer in Au₂Ag₈Cu₅ are slightly longer than those of Ag₉Cu₆ (e.g. M_octahedron–C_ligand: 1.886 Å vs. 1.870 Å; M_cube–C_ligand: 2.394 Å vs. 2.381 Å), further attesting the similarity of the outer layer between Au₂Ag₈Cu₅ and Ag₉Cu₆.

Fig. 3 Structural analysis in body-centered cubic (BCC) M₁₅ NCs. (a) Overall and space-filling structure of monocationic Au₂Ag₈Cu₅, (b) the five types of linear ^tBu-C≡C–CuC≡C-^tBu staple motifs on the metal surface. Coordination modes of ^tBu-C≡C ligands: μ₂-η₁ (Ag/Cu), η₁ (Ag); μ₂-η₁ (Au/Ag/Cu), η₁ (Ag); μ₂-η₁ (Au/Ag/Cu), η₁ (Cu) and μ₂-η₁ (Ag/Cu), η₁ (Cu). (c) Structural analysis and (d) anatomy of BCC M₁₅ kernel in Au₂Ag₈Cu₅, Ag₉Cu₆, and Au₇Ag₈, respectively. Color legend: Au, yellow; Ag, cyan; Cu, orange; C, gray; H, white.

Electrocatalytic CO₂ reduction performance of the three M₁₅ NCs in flow cell

As the three NCs possess a M₁₅ configuration yet different metal core and significant discrepancies are observed on the physiochemical properties (e.g. optical absorbance), we wonder whether they have different catalytic properties. To probe the metal core effect, we next examined the electrocatalytic performance of the three catalysts on gas diffusion electrode (GDL, 2 × 1.5 cm²) toward CO₂RR by constant-potential electrolysis (CPE) measurements at various applied potentials in a custom-designed flowcell (Fig. 4a) (the electrochemical measurement details can be found in ESI†). The linear scanning voltammetry (LSV) was first conducted for Au₇Ag₈/GDL, Ag₉Cu₆/GDL, and Au₂Ag₈Cu₅/GDL. As depicted in Fig. S10,† for all the samples, a sudden decrease in the reduction current can be observed after the first potential sweep (black line) along with the onset potential shifted positively. In the second and third sweeps (red and blue line), the resulting current and onset potential remained unchanged. It suggests that, all the catalysts were activated, and such ligand stripping phenomenon has been recorded in thiolate-protected Au₂₅ NCs in the CO₂ electroreduction process as well.⁴⁶

Fig. 4 (a) Exploded diagrams of the electrochemical reactors for CO₂ electroreduction in flow cell. (b) CO and (c) formate faradaic efficiency for Au₇Ag₈, Ag₉Cu₆, and Au₂Ag₈Cu₅ NC/GDLs examined at different applied potentials. (d) FEs for various CO₂RR products obtained on the three NC/GDLs. The corresponding (e) CO and (f) formate partial current density. (g) Long-term stability of Au₇Ag₈/GDL, Ag₉Cu₆/GDL, and Au₂Ag₈Cu₅/GDL at −0.49 V, −1.19 V, and −0.99 V (vs. RHE), respectively. (top) i–t curve; (bottom) FEs of CO, H₂, and formate at different time.

For all the catalysts, CO is the main product at more positive potentials, and formate at more negative potentials was also detected in the liquid phase by ¹H-NMR for the two Cu-containing catalysts (Fig. S11 and S12†). As shown in Fig. 4b, Au₇Ag₈ exhibited high selectivity for CO formation, evidenced by the higher FE_CO values at all tested potentials, ranging from ∼86.6% at −0.39 V to ∼98.1% at −1.19 V (vs. RHE). In contrast, for both Ag₉Cu₆ and Au₂Ag₈Cu₅, a similar volcanic shape on the FE_CO value is observed, in which the highest FE_CO value of ∼94.2% and ∼95.0% at −0.49 V is obtained for Ag₉Cu₆ and Au₂Ag₈Cu₅, respectively. However, for these two NCs, CO has higher FEs at more positive potentials, whereas the FE for formate (FE_formate) increases rapidly when the potential goes more negatively (Fig. 4c). The largest FE_formate value for Ag₉Cu₆ and Au₂Ag₈Cu₅ is 47.0% at −1.19 V and 28.3% at −0.99 V, respectively. However, the FE_formate value for Au₂Ag₈Cu₅ decreased to less than 20% at −1.19 V. Impressively, for both bimetallic NCs of Ag₉Cu₆ and Au₇Ag₈, the H₂ evolution can be significantly suppressed, as the FE_H2 is less than 10% in the whole tested potential range (Fig. S13a†). However, for Au₂Ag₈Cu₅, when the potential goes more negatively, the HER becomes more dominant, and the highest FE_H2 can reach 37.0% at −1.19 V. The total FE values of the products for the three catalysts were presented as a function of applied potential from −0.59 V to −1.19 V, showing that CO, H₂, and formate are the main products with a total FE value close to 100% in the whole potential range (Fig. 4d). No other product was detected by NMR or GC. By deducting H₂, at −0.49 V, the highest FE_CO for Au₇Ag₈ is ∼98.1%, while the highest FE_CO+formate for Ag₉Cu₆ and Au₂Ag₈Cu₅ is ∼100.0% and ∼97.4%, respectively, indicating the three NCs can efficiently convert CO₂ into value-added carbon products (Fig. 4d). Meanwhile, the CO partial current density (j_CO) increased with the increasing of applied potential for the Au₇Ag₈ and Ag₉Cu₆ catalysts, and j_CO reached the maximal value at −0.99 V then diminished at −1.19 V for Au₂Ag₈Cu₅ (Fig. 4e). Au₇Ag₈ had a much larger j_CO value than Ag₉Cu₆ and Au₂Ag₈Cu₅ at all potentials, further manifesting its unique advantage for converting CO₂ into CO exclusively. Furthermore, the partial current density of formate (j_formate) for Ag₉Cu₆ and Au₂Ag₈Cu₅ exhibited the same trend with the FE_formate in the tested potential range (Fig. 4f), that is, j_formate increased from −0.39 V to −1.19 V, reaching the maximal value of 49.1 mA cm⁻² at −1.19 V for Ag₉Cu₆, however, for Au₂Ag₈Cu₅, it first increased then decreased, and the maximal value is 32.7 mA cm⁻² at −0.99 V. This is mainly due to that the HER process became dominant at very high negative potentials (Fig. S13b†).

Stability is another important criterion to evaluate the catalytic property of the electrocatalyst, hence the long-term stability of Au₇Ag₈, Ag₉Cu₆, and Au₂Ag₈Cu₅ was tested at −0.49 V, −1.19 V, and −0.99 V, respectively. As illustrated in Fig. 4g, the current density and corresponding FE value of Au₇Ag₈ and Ag₉Cu₆ remained almost unchanged after 10 h's continuous operation, indicating robust long-term durability, however, under the same conditions, the current density of Au₂Ag₈Cu₅ decreased about 15% (from 75.4 to 64.1 mA cm⁻²), meanwhile the FE_CO+formate decreased and FE_H2 increased gradually, presumably due to that the asymmetric metal core of Au₂Ag₈Cu₅ is easier to decompose and/or aggregate during the electrocatalytic process. We conducted the MS measurement of the three NCs before and after CO₂RR to examine the change. As shown in Fig. S14,† for the Cu-containing NCs, the molecular ions with strong signal (m/z = 2548.6640 for Au₂Ag₈Cu₅, m/z = 2325.5677 for Ag₉Cu₆) are still dominant, indicating both clusters are rather robust. However, for both clusters, some peaks in the lower m/z region appeared, suggesting some of the cluster molecules decomposed. For Au₂Ag₈Cu₅, after CO₂RR, two peaks with m/z = 1112.4193 (Fragment A) and m/z = 1471.6617 (Fragment B) can be assigned to Au₄L₄ (L: C₆H₉, cal. MW: 1112.4183) and Au₅L₆⁺ (cal. MW: 1471.6608), respectively. In addition, compared to product A, the peak intensity of Au₂Ag₈Cu₅ decreased, suggesting that some Au₂Ag₈Cu₅ molecules might decompose to Au-alkynyl complexes and/or metal nanoparticles; for Ag₉Cu₆, after CO₂RR, there are three peaks appearing at m/z = 1020.8771 (Fragment A), 1099.9494 (Fragment B), and 1183.0180 (Fragment C), which can be assigned to Ag₂Cu₅L₆⁺ (cal. MW: 1020.8762), Ag₂Cu₅L₇⁺ (cal. MW: 1099.9485), and Ag₂Cu₅L₈⁺ (cal. MW: 1183.0171), respectively. Also, there is one peak with m/z at 2476.3047 (D), which can be assigned to Ag₁₂Cu₂L₁₃⁺ (cal. MW: 2476.3039), and it is probably formed in the chamber during the MS measurement. For Au₇Ag₈, after CO₂RR, there are two minor peaks appearing at m/z = 3053.0904 (Fragment A) and 2937.2619 (Fragment B), which can be assigned to Au₇Ag₈L₁₀⁺ (cal. MW: 3053.0910) and Au₆Ag₈L₁₁⁺ (cal. MW: 2937.2625), respectively. These results indicate that, the majority of the cluster molecules can be well preserved during the CO₂RR process.

Furthermore, we also tested the recover capability of the Ag₉Cu₆ and Au₂Ag₈Cu₅ catalysts for CO₂RR. Using the 579 and 484 nm fingerprint absorbance peak as the metric, the absorbance change can be quantified and employed to estimate the recovery rate (Fig. S15 and S16†). It is worth noting that, besides the intensity of the characteristic peak decreased with different extents at different potentials, the whole absorbance feature of these two NCs remained intact. The calculated results are summarized in Tables S9 and S10.† Specifically, from −0.39 V to −1.19 V, the intensity of the absorbance peak at 579 nm decreased gradually, and the recovery rate of these two NCs decreased as well. Also, the recovery rate of Ag₉Cu₆ ranges from 30.4% to 96.6%, higher than that of Au₂Ag₈Cu₅ (21.0% to 89.7%) at all applied potentials, in good agreement with the finding in the i–t test.

The observation that all M₁₅ NCs exhibited high catalytic selectivity of CO₂ electroreduction to CO at the low potentials and that the clusters containing Cu metals, namely Ag₉Cu₆ and Au₂Ag₈Cu₅, were found to generate formate products is interesting. We then compared the formation selectivity of formate and CO of the three M₁₅ NCs with recently reported atomically precise metal nanoclusters in CO₂RR, as summarized in Table S11 and S12†, respectively. Although the reports were conducted in different cell type such as H-cell, flow-cell and MEA-cell, it can be noted that, the formate selectivity of the Ag₉Cu₆ and Au₂Ag₈Cu₅ clusters is lower than the Cu₃₂ cluster, but much higher than the Au₂₅ cluster and all the AuCd alloy clusters. For CO formation selectivity, the highest FE_CO value of the Au₇Ag₈, Ag₉Cu₆, and Au₂Ag₈Cu₅ clusters are all over 94%, at least comparable with, if not superior to, the Au, Ag clusters and the AuCd, AuPd, AuAg alloy clusters. Particularly, the FE_CO value of the Au₇Ag₈ cluster can reach as high as 98.1%, larger than most of the recent reports, quite close to the Au₂₅(PET)₁₈ and Au₂₄Pd₁(PET)₁₈ clusters (∼100% for both).

CO₂RR mechanistic study by DFT calculations

To deeply comprehend the electrocatalytic mechanism, we next performed DFT calculations (see ESI† for computational details) to determine the optimal catalytic site and analyze the selectivity difference. To simplify the calculation, all –C≡C-^tBu ligands are replaced with –C≡C-CH₃. The optimized structure based on the crystal structure of Au@AuAg₄Cu₃@Ag₄Cu₂ is used as a model for DFT calculation, as shown in Fig. S17.† On the intact [Au₇Ag₈(C≡C-CH₃)₁₂]⁺ (Fig. S18a†) and [Ag₉Cu₆(C≡C-CH₃)₁₂]⁺ (Fig. S18b†), CO₂RR and HER compete on the same staple metal site (Au for [Au₇Ag₈(C≡C-CH₃)₁₂]⁺ and Cu for [Ag₉Cu₆(C≡C-CH₃)₁₂]⁺). While for [Au₂Ag₈Cu₅(C≡C-CH₃)₁₂]⁺ NC, the staple Cu acts as the active site for HCOO* binding, while *COOH, *CO, and *H tend to bond with the sub-surface Au atom (Fig. S18c†). The free energy diagrams of CO₂RR and HER on these three intact systems are depicted in Fig. 5a, c, and e, and the H₂ pathway is thermodynamically more favourable than CO₂RR. In addition, we found that the bonding of *H on clusters containing copper is stronger. To uncover this phenomenon, we performed Bader charge analysis of metal active site (Table S13†), which intuitively shows that *H on Ag₉Cu₆ has the most negative charge (−0.29 |e|), indicating a stronger adsorption. However, *H has the strongest bonding on Au₂Ag₈Cu₅ cluster, which is possibly due to the special coordination environment of active Au atom, so that Au with greater electronegativity can rob electrons from the surrounding Ag or Cu. Therefore, the active Au here is negatively charged (−0.15 |e|) and interacts strongly with *H.

Fig. 5 (a, c and e) Comparison of CO₂RR vs. HER on three intact NCs. (b, d and f) Reaction scheme for CO₂ electroreduction on single ligand-removed clusters to form CO via the proton mechanism (blue region) and to form formate via the hydride–proton mechanism (green region) at zero applied potential. The reaction step with the highest free-energy change step is framed in red.

Inspired by related studies^17,19,46 and our recent finding on Ag₁₅ NC for CO₂RR,³³ ligand removal to expose undercoordinated metal atom may serve as the electrocatalytic active centers. As depicted in Fig. S19,† the release of one –C≡C-CH₃ results in exposure of four shell–metal atoms to form two (111)-like triangular faces. Due to the highly symmetrical structures of [Ag₉Cu₆(C≡C-CH₃)₁₂]⁺ and [Au₂Ag₈Cu₅(C≡C-CH₃)₁₂]⁺, the removal of either –C≡C-CH₃ ligand is equivalent. However, it is predicted that, the removal of alkynyl ligand bonded to two Ag atoms near the shell Au atom on the [Au₂Ag₈Cu₅(C≡C-CH₃)₁₂]⁺ cluster is more thermodynamically supported (the red circle marked in Fig. S19c†), thereby exposing 111-like surfaces (Fig. S19d†). In this context, the H* would readily adsorb to the hollow position of the triangle in a bridging manner (Fig. S20†). The calculated Bader charge (Fig. S20†) shows that the adsorbed *H has a negative charge of −0.11 to −0.22 |e|, suggesting that the adsorbed *H functions as a hydride and may provide the hydrogen source for CO₂ reduction.^24,47 As a consequence, there are four possible reaction channels: (1) proton mechanism of reacting CO₂ with the proton from solution; (2) hydride mechanism of reacting CO₂ with the capping hydride (H*); (3) the hydride–proton or (4) proton–hydride mechanism, where hydride and proton alternately participate in the catalytic process. The free energy difference (ΔG) of each reduction step can be found in Fig. S21 to S23.† On the ligand-removed NCs the proton-reduction channel is preferred for CO; whereas the hydride–proton channel is more favoured for formate, that is, the first adsorbed H* (marked in green) is easily transferred to the C atom to form HCOO*, the second adsorbed H*(marked in blue) is difficult to transfer, but can occupy the active site to facilitate subsequent protonation. The overall mechanism of CO formation via the proton mechanism and formate formation via the hydride–proton mechanism from CO₂ reduction on three NCs are summarized in Fig. S24 to S26.† The corresponding free energy profile for generating CO and formate are shown in Fig. 5b, d, and f. Apparently, for the CO pathway, the formation of *COOH is the potential-determining step (PDS), the same as on intact NCs; for the formate pathway, the PDS corresponds to the electrochemical protonation of *HCOO to formate or the transfer of H* to the C atom to form HCOO*. Their comparable ΔG for PDS (CO is slightly preferred) indicates that the CO and formate formation is competitive, which is consistent with the experimental observation that CO and formate are the main products on Ag₉Cu₆ and Au₂Ag₈Cu₅. The corresponding optimal configurations of key intermediates are depicted in Fig. 6. On both [Ag₉Cu₆(C≡C-CH₃)₁₁]⁺ and [Au₂Ag₈Cu₅(C≡C-CH₃)₁₁]⁺, the two O atoms of HCOO* bind tightly with one Cu atom and one Ag atom on the metal triangle. The active site for CO formation differs from each other, where the trans-COOH* prefers to bind to Ag atom on [Au₇Ag₈(C≡C-CH₃)₁₁]⁺, to Cu atom on [Ag₉Cu₆(C≡C-CH₃)₁₁]⁺, and to Au atom on [Au₂Ag₈Cu₅(C≡C-CH₃)₁₁]⁺. The H* on all three clusters easily occupy the hollow sites of the triangle in a bridging manner. Note that, the attraction between the negatively charged H* and the positively charged C of the CO₂ reactant can trigger the favourable H* transfer to form *HCOO, and the participation of metallic Cu as the active center is also important in stabilizing the *HCOO intermediate. Based on the high CO and formate selectivity observed in experiments, the exposure of more active surface metal site upon ligand removal could be the real reason for the feasible CO₂RR pathway. It is worth noting that we use a simplified –C≡C-CH₃ group for our simulation, while in experiment, much bulkier butyl groups are employed for protection. To further illustrate the feasibility of this simplification, we investigated the CO₂RR and HER performance of [Ag₉Cu₆(C≡C-^tBu)₁₂]⁺ synthesized in the actual experiment and compared it with [Ag₉Cu₆(C≡C-CH₃)₁₂]⁺. As shown in Fig. S27,† the bulkiness brought by –C≡C-^tBu groups slightly weakens the adsorption strength for intermediate state. However, the predicted response and the PDS are basically the same. Thus, the simplification of the butyl ligand can provide valid prediction on the performance.

Fig. 6 Schematic presentation of key intermediates on (a) [Au₇Ag₈(C≡C-CH₃)₁₁]⁺, (b) [Ag₉Cu₆(C≡C-CH₃)₁₁]⁺, and (c) [Au₂Ag₈Cu₅(C≡C-CH₃)₁₁]⁺ NCs, respectively. Color legend: Au, gold; Ag, blue; Cu, brick-red; C, gray; O, red; H, white (mark the first H* in green and the second H* in blue).

Discussion on metal core effect of the M₁₅ NCs toward CO₂RR

Finally, with the combined experimental and theoretical results of the three M₁₅ clusters for CO₂RR, plus the reported Ag₁₅ one,³³ we would like to discuss the metal core effect of the M₁₅ series toward CO₂RR. Note that, Ag₁₅ and Au₇Ag₈ clusters can exclusively convert CO₂ into CO with very high FE values, however, for Au₂Ag₈Cu₅ and Ag₉Cu₆ clusters, formate can be generated. Apparently, the presence of Cu atoms is critical for generating formate, and more importantly, with two-atom difference (Au₂Ag₈Cu₅vs. Ag₉Cu₆), the catalytic performance is drastically different (the FE_formate value for Ag₉Cu₆ is higher than that of Au₂Ag₈Cu₅, and the latter one has stronger H₂ evolution than formate formation at −1.19 V). That means, strong core size effect toward CO₂RR is observed in the M₁₅ series, and metal exchange is an effective strategy to fine-tune the electrocatalytic performance of homoleptic alkynyl-protected metal nanoclusters. The theoretical calculations also support the above findings. For both Ag₁₅ and Au₇Ag₈ clusters, the undercoordinated Ag and/or Au atoms upon one intact ligand stripping are the active sites for CO formation. However, for Ag₉Cu₆ and Au₂Ag₈Cu₅ NCs, the undercoordinated Cu atoms can serve as the active sites for CO and formate formation. As revealed from Fig. S19,† the different electrocatalytic performance can be ascribed not only from the core atom difference (Ag vs. Au), but more importantly, the exposed (111)-like Ag₂Cu₂ surface and Au₁Cu₁Ag₂ surface for Ag₉Cu₆ and Au₂Ag₈Cu₅, respectively.

Conclusions

In conclusion, the first all-alkynyl-protected trimetallic superatom of Au₂Ag₈Cu₅ is synthesized through a metal exchange approach, of which the formation process is elucidated. Au₂Ag₈Cu₅ NC has a similar M@M₈@M₆ metal core configuration with Ag₉Cu₆ and Au₇Ag₈ NCs, but quite different absorbance feature. Moreover, the three NCs exhibited drastically different catalytic performance toward CO₂RR, in which Au₇Ag₈ can convert CO₂ into CO exclusively, while CO and formate are the main products for Ag₉Cu₆ and Au₂Ag₈Cu₅ at more negative potentials with the highest FE_formate of 47.0% and 28.3%, respectively. DFT calculations revealed that ligand stripping can expose more active surface metal atoms to boost CO₂RR activity and selectivity. The formation of surface hydride plays a critical role in triggering the formation and stabilization of HCOO* on the Ag–Cu active center, leading to the exclusive formation of formate in the Cu-containing NCs. Strong core effect toward CO₂RR is observed. This study not only provides an ingenious strategy to tailor the metal core of alkynyl-protected metal NCs at atomic level, but also highlights the unique advantages of employing metal NCs as model catalysts to advance the fundamental mechanistic understanding toward CO₂RR and beyond.

Data availability

All the data in this study are provided in the main text and ESI.†

Author contributions

Z. T. conceived the idea, X. M. conducted most of the experiments, L. Q. and Y. L. helped the characterization, F. S. and Q. T. conducted the DFT calculations, X. K. and L. W. provided the technique support for CO₂RR test, D. J. offered guidance for theoretical calculations, X. M. and Z. T. wrote up the draft, Q. T. and Z. T. provided the funding support, and all the authors contributed to the final proof of the manuscript.

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

This study is supported by the Open Fund of Guangdong Provincial Key Laboratory of Functional Supramolecular Coordination Materials and Applications (No. 2021A07). Z. T. acknowledges the financial support from Guangdong Natural Science Funds (No. 2022A1515011840). L. W. acknowledges the funding from National Natural Science Foundation of China (No. 21805170). Q. T. thanks the grants from the National Natural Science Foundation of China (No. 21903008) and the Chongqing Science and Technology Commission (cstc2020jcyj-msxmX0382).

Notes and references

1. Q. Lu and F. Jiao, Nano Energy, 2016, 29, 439–456.
2. Y. Wu, Z. Jiang, X. Lu, Y. Liang and H. Wang, Nature, 2019, 575, 639–642.
3. H. B. Yang, S.-F. Hung, S. Liu, K. Yuan, S. Miao, L. Zhang, X. Huang, H.-Y. Wang, W. Cai, R. Chen, J. Gao, X. Yang, W. Chen, Y. Huang, H. M. Chen, C. M. Li, T. Zhang and B. Liu, Nat. Energy, 2018, 3, 140–147.
4. Y. Quan, J. Zhu and G. Zheng, Small Sci., 2021, 1, 2100043.
5. S. Nitopi, E. Bertheussen, S. B. Scott, X. Liu, A. K. Engstfeld, S. Horch, B. Seger, I. E. L. Stephens, K. Chan, C. Hahn, J. K. Nørskov, T. F. Jaramillo and I. Chorkendorff, Chem. Rev., 2019, 119, 7610–7672.
6. S. Jin, Z. Hao, K. Zhang, Z. Yan and J. Chen, Angew. Chem., Int. Ed., 2021, 60, 20627–20648.
7. K. D. Gilroy, A. Ruditskiy, H. C. Peng, D. Qin and Y. Xia, Chem. Rev., 2016, 116, 10414–10472.
8. S. Zhang, Q. Fan, R. Xia and T. J. Meyer, Acc. Chem. Res., 2020, 53, 255–264.
9. L. Qin, G. Ma, L. Wang and Z. Tang, J. Energy Chem., 2021, 57, 359–370.
10. S. Li, A. V. Nagarajan, Y. Li, D. R. Kauffman, G. Mpourmpakis and R. Jin, Nanoscale, 2021, 13, 2333–2337.
11. D. R. Kauffman, D. Alfonso, C. Matranga, H. Qian and R. Jin, J. Am. Chem. Soc., 2012, 134, 10237–10243.
12. S. Zhao, N. Austin, M. Li, Y. B. Song, S. D. House, S. Bernhard, J. C. Yang, G. Mpourmpakis and R. C. Jin, ACS Catal., 2018, 8, 4996–5001.
13. B. Kumar, T. Kawawaki, N. Shimizu, Y. Imai, D. Suzuki, S. Hossain, L. V. Nair and Y. Negishi, Nanoscale, 2020, 12, 9969–9979.
14. X. Lin, W. Ma, K. Sun, B. Sun, X. Fu, X. Ren, C. Liu and J. Huang, J. Phys. Chem. Lett., 2020, 12, 552–557.
15. N. Austin, S. Zhao, J. R. McKone, R. Jin and G. Mpourmpakis, Catal. Sci. Technol., 2018, 8, 3795–3805.
16. D. R. Alfonso, D. Kauffman and C. Matranga, J. Chem. Phys., 2016, 144, 184705.
17. S. Li, D. Alfonso, A. V. Nagarajan, S. D. House, J. C. Yang, D. R. Kauffman, G. Mpourmpakis and R. Jin, ACS Catal., 2020, 10, 12011–12016.
18. S. Zhuang, D. Chen, L. Liao, Y. Zhao, N. Xia, W. Zhang, C. Wang, J. Yang and Z. Wu, Angew. Chem., Int. Ed., 2020, 59, 3073–3077.
19. S. Li, A. V. Nagarajan, D. R. Alfonso, M. Sun, D. R. Kauffman, G. Mpourmpakis and R. Jin, Angew. Chem., Int. Ed., 2021, 60, 6351–6356.
20. Y. N. Sun, X. Liu, K. Xiao, Y. Zhu and M. Y. Chen, ACS Catal., 2021, 11, 11551–11560.
21. Z. X. Tao, Z. S. Wu, Y. S. Wu and H. L. Wang, ACS Catal., 2020, 10, 9271–9275.
22. Z. Gu, H. Shen, Z. Chen, Y. Yang, C. Yang, Y. Ji, Y. Wang, C. Zhu, J. Liu, J. Li, T.-K. Sham, X. Xu and G. Zheng, Joule, 2021, 5, 429–440.
23. Y. F. Wang, Z. Chen, P. Han, Y. H. Du, Z. X. Gu, X. Xu and G. F. Zheng, ACS Catal., 2018, 8, 7113–7119.
24. Q. Tang, Y. Lee, D. Y. Li, W. Choi, C. W. Liu, D. Lee and D. E. Jiang, J. Am. Chem. Soc., 2017, 139, 9728–9736.
25. Z. Lei, X. K. Wan, S. F. Yuan, Z. J. Guan and Q. M. Wang, Acc. Chem. Res., 2018, 51, 2465–2474.
26. Z. Lei and Q.-M. Wang, Coord. Chem. Rev., 2019, 378, 382–394.
27. P. Maity, S. Takano, S. Yamazoe, T. Wakabayashi and T. Tsukuda, J. Am. Chem. Soc., 2013, 135, 9450–9457.
28. X. Ma, Y. Tang, G. Ma, L. Qin and Z. Tang, Nanoscale, 2021, 13, 602–614.
29. Z. Lei, X. K. Wan, S. F. Yuan, J. Q. Wang and Q. M. Wang, Dalton Trans., 2017, 46, 3427–3434.
30. P. Yuan, R. Zhang, E. Selenius, P. Ruan, Y. Yao, Y. Zhou, S. Malola, H. Hakkinen, B. K. Teo, Y. Cao and N. Zheng, Nat. Commun., 2020, 11, 2229.
31. Z. Y. Wang, M. Q. Wang, Y. L. Li, P. Luo, T. T. Jia, R. W. Huang, S. Q. Zang and T. C. W. Mak, J. Am. Chem. Soc., 2018, 140, 1069–1076.
32. T. T. Jia, G. Yang, S. J. Mo, Z. Y. Wang, B. J. Li, W. Ma, Y. X. Guo, X. Chen, X. Zhao, J. Q. Liu and S. Q. Zang, ACS Nano, 2019, 13, 8320–8328.
33. L. Qin, F. Sun, X. Ma, G. Ma, Y. Tang, L. Wang, Q. Tang, R. Jin and Z. Tang, Angew. Chem., Int. Ed., 2021, 60, 26136–26141.
34. X. Ma, L. Xiong, L. Qin, Y. Tang, G. Ma, Y. Pei and Z. Tang, Chem. Sci., 2021, 12, 12819–12826.
35. X. Kang, X. Wei, X. Liu, S. Wang, T. Yao, S. Wang and M. Zhu, Nat. Commun., 2021, 12, 6186.
36. H. Shen, Z. Xu, L. Wang, Y. Z. Han, X. Liu, S. Malola, B. K. Teo, H. Hakkinen and N. Zheng, Angew. Chem., Int. Ed., 2021, 60, 22411–22416.
37. Y. Wang, H. Su, L. Ren, S. Malola, S. Lin, B. K. Teo, H. Hakkinen and N. Zheng, Angew. Chem., Int. Ed., 2016, 55, 15152–15156.
38. Z. Tang, D. A. Robinson, N. Bokossa, B. Xu, S. Wang and G. Wang, J. Am. Chem. Soc., 2011, 133, 16037–16044.
39. W. Du, S. Jin, L. Xiong, M. Chen, J. Zhang, X. Zou, Y. Pei, S. Wang and M. Zhu, J. Am. Chem. Soc., 2017, 139, 1618–1624.
40. X. Zou, Y. Li, S. Jin, X. Kang, X. Wei, S. Wang, X. Meng and M. Zhu, J. Phys. Chem. Lett., 2020, 11, 2272–2276.
41. T. Higaki, Q. Li, M. Zhou, S. Zhao, Y. Li, S. Li and R. Jin, Acc. Chem. Res., 2018, 51, 2764–2773.
42. Q. Li, T. Y. Luo, M. G. Taylor, S. Wang, X. Zhu, Y. Song, G. Mpourmpakis, N. L. Rosi and R. Jin, Sci. Adv., 2017, 3, e1603193.
43. A. Ghosh, O. F. Mohammed and O. M. Bakr, Acc. Chem. Res., 2018, 51, 3094–3103.
44. X. Kang, X. Wei, S. Jin, Q. Yuan, X. Luan, Y. Pei, S. Wang, M. Zhu and R. Jin, Proc. Natl. Acad. Sci., 2019, 116, 18834–18840.
45. K. Zheng, V. Fung, X. Yuan, D. E. Jiang and J. Xie, J. Am. Chem. Soc., 2019, 141, 18977–18983.
46. H. Seong, V. Efremov, G. Park, H. Kim, J. S. Yoo and D. Lee, Angew. Chem., Int. Ed., 2021, 60, 14563–14570.
47. F. Li and Q. Tang, J. Catal., 2020, 387, 95–101.

---

Footnotes

† Electronic supplementary information (ESI) available: synthesis, characterization, supporting figures and tables. Details and crystal data of [Au₂Ag₈Cu₅(C≡C-^tBu)₁₂]SbF₆ (CIF). CCDC 2072663. The videos for the metal exchange process to synthesize Au₂Ag₈Cu₅ from Ag₉Cu₆. For ESI and crystallographic data in CIF or other electronic format see https://doi.org/10.1039/d2sc02886g

‡ X. Ma and F. Sun contributed equally to this work.

---