The role of metal accessibility on carbon dioxide electroreduction in atomically precise nanoclusters

Atomically precise nanoclusters (NCs) can be designed with high faradaic efficiency for the electrochemical reduction of CO2 to CO (FECO) and provide useful model systems for studying the metal-catalysed CO2 reduction reaction (CO2RR). While size-dependent trends are commonly evoked, the effect of NC size on catalytic activity is often convoluted by other factors such as changes to surface structure, ligand density, and electronic structure, which makes it challenging to establish rigorous structure–property relationships. Herein, we report a detailed investigation of a series of NCs [AunAg46−n(C 
<svg xmlns="http://www.w3.org/2000/svg" version="1.0" width="23.636364pt" height="16.000000pt" viewBox="0 0 23.636364 16.000000" preserveAspectRatio="xMidYMid meet"><metadata>
Created by potrace 1.16, written by Peter Selinger 2001-2019
</metadata><g transform="translate(1.000000,15.000000) scale(0.015909,-0.015909)" fill="currentColor" stroke="none"><path d="M80 600 l0 -40 600 0 600 0 0 40 0 40 -600 0 -600 0 0 -40z M80 440 l0 -40 600 0 600 0 0 40 0 40 -600 0 -600 0 0 -40z M80 280 l0 -40 600 0 600 0 0 40 0 40 -600 0 -600 0 0 -40z"/></g></svg>
 CR)24Cl4(PPh3)2, Au24Ag20(CCR)24Cl2, and Au43(CCR)20/Au42Ag1(CCR)20] with similar sizes and core structures but different ligand packing densities to investigate how the number of accessible metal sites impacts CO2RR activity and selectivity. We develop a simple method to determine the number of CO2-accessible sites for a given NC then use this to probe relationships between surface accessibility and CO2RR performance for atomically precise NC catalysts. Specifically, the NCs with the highest number of accessible metal sites [Au43(CCR)20 and Au42Ag1(CCR)20] feature a FECO of >90% at –0.57 V vs. the reversible hydrogen electrode (RHE), while NCs with lower numbers of accessible metal sites have a reduced FECO. In addition, CO2RR studies performed on other Au–alkynyl NCs that span a wider range of sizes further support the relationship between FECO and the number of accessible metal sites, regardless of NC size. This work establishes a generalizable approach to evaluating the potential of atomically precise NCs for electrocatalysis.


Introduction
5][6] Gold-and silver-based NPs are particularly effective for the selective reduction of CO 2 to CO. 7 Though the effects of nanoparticle size, 8-10 shape, 11,12 and surface ligands 13,14 on the CO 2 reduction reaction (CO 2 RR) have been widely studied, the nonuniformity of metal NP catalysts is a long-standing challenge in the investigation of fundamental catalytic mechanisms. 1,15In particular, it is oen difficult to identify the specic active sites that drive catalysis because of the wide distribution of local microenvironments in ligandprotected NPs that adopt varying sizes, shapes, and surface structures. 16,17For example, although functionalization with larger organic ligands has been shown to enhance the CO 2 RR activity of Au NPs, 18,19 uncertainty over the exact arrangement of surface ligands makes it difficult to determine how bulky ligands impact selectivity and catalytic activity.Such molecularlevel insight is, however, possible when atomically precise nanoclusters (NCs) are used as catalysts, since their uniformity allows the entire particle structure-including the ligand shell-to be resolved crystallographically. [20][21][22][23] Soon aer the canonical Au 25 (SR) 18 (SR = aryl or alkylthiolate) NC was rst reported, it was shown to be effective for CO 2 RR, featuring a high faradaic efficiency for CO (FE CO ) at -1 V vs. the reversible hydrogen electrode (RHE). 24Though the CO 2 RR has since been studied for many other atomically precise Au NCs, [25][26][27][28] there still remains much to be understood about how NC size, structure, and surface ligand identity inuence catalytic activity and selectivity.For instance, relationships between NC size and CO 2 RR activity are challenging to identify because the ligand-to-metal ratio typically increases for smaller NCs, [29][30][31] resulting in higher surface coverage.Changes to the arrangement of surface ligands and metal atoms-as well as the electronic structure of the NC-may also affect the outcome of catalytic reactions, further convoluting structure-property relationships. 32Indeed, differing trends have been reported for how NC size affects CO 2 RR activity.For example, in the series Au 25 (SR) 18 , Au 38 (SR) 24 , and Au 144 (SR) 60 (SR = SC 2 H 4 Ph), CO 2 RR activity increases with increasing NC size, 33 while other studies have found that the FE CO of Au-SR NCs is not directly affected by NC size. 34Given the different size-dependent trends that have been observed for atomically precise NCs, the number of active sites is oen a better predictor of catalytic behavior but is difficult to manipulate in a predictable fashion.
In an effort to decouple the role of metal active sites from NC size, structure, and ligand type, we designed a series of alkynylprotected atomically precise Au/Ag NCs with similar sizes and core structures but different degrees of surface ligand coverage.We investigated the CO 2 RR performance of these NCs and developed a convenient computational method to quantitatively evaluate the accessibility of potential catalytically active sites.Critically, the use of acetylene-based ligands-bearing one rotatable bond-simplies the conformational landscape at the metal-ligand interface, thereby clarifying the effect of ligand modication on surface coverage and the number of accessible metal sites.In particular, we found that the number and accessibility of surface metal sites is directly correlated to experimental CO 2 RR activity.

Synthesis and characterization of alkynyl-protected Au/Ag NCs
We recently reported the isostructural alkynyl-protected NCs Au 43 (C^C t Bu) 20 (Au 43 ) and Au 42 Ag 1 (C^C t Bu) 20 (Au 42 Ag 1 ), which are synthesized by reducing an oligomeric Au I -C^C t Bu or Au I /Ag I -C^C t Bu precursor with borane tert-butylamine then purifying via thin layer chromatography (Fig. S1A †). 35With a nearly identical NC core but an increased density of alkynyl surface ligands, we also selected the previously reported Au 24 -Ag 20 (C^CPh t Bu) 24 Cl 2 NC (Au 24 Ag 20 ) for comparison (Fig. S1B †). 36,37To complete a series of NCs with similar sizes and varying surface ligand densities, we also targeted an Au/Agalkynyl NC with an even denser organic shell.This was achieved by introducing a bulky triphenylphosphine (PPh 3 ) ligand through the "hydride-mediated conversion" method. 38,39pecically, [Au 9 (PPh 3 ) 8 ] 3+ (ref.40) was reduced with NaBH 4 to furnish a hydride-doped [HAu 9 (PPh 3 ) 8 ] 2+ cluster, which was then reacted with CH 3 COOAg, meta-substituted phenylacetylene ligands, and triethylamine to yield NCs with a composition of Au n Ag 46−n (C^CPh-m-X) 24 Cl 4 (PPh 3 ) 2 (Au n -Ag 46−n , n = 16-19, X = H, F, CH 3 ) (Fig. 1, see ESI † for experimental details).Though attempts to grow single crystals of Au n Ag 46−n (C^CPh) 24 Cl 4 (PPh 3 ) 2 suitable for structure determination were unsuccessful, crystal structures were successfully determined for Au n Ag 46−n with X = F and CH 3 (see Fig. 2).The successful crystallization of Au n Ag 46−n with meta-substituted phenylacetylene ligands can be attributed to the additional interparticle C-H/p, or C-F/p interactions between the meta functional groups on one NC and the phenyl rings of ligands on another NC (Fig. S2 and S3 †). 41,42e solution-phase UV-vis absorption spectra of all three Au n Ag 46−n NCs with different protecting ligands exhibit sharp absorption peaks centered near 400 and 690 nm (Fig. 1A).Moreover, electrospray ionization mass spectrometry (ESI-MS) analysis shows 2+ and 1+ ion peaks for the NCs (Fig. 1B).Combined with the X-ray crystallography data, the 2+ and 1+ charges are attributed to ionization in ESI-MS-not the native charge states of the NCs-and the peaks correspond to NCs that have lost one or two PPh 3 ligands, which has been commonly observed for similar atomically precise NCs (Fig. S4-S6 †). 43,44he ESI-MS spectrum of Au n Ag 46−n (C^CPh-m-F) 24 Cl 4 (PPh 3 ) 2 contains peaks corresponding to n = 16, 17, and 18 (Fig. S5 †), consistent with the crystallographically rened composition of Au 17.67 Ag 28.33 (C^CPh-m-F) 24 Cl 4 (PPh 3 ) 2 (Tables S1 and S2 †).When HC^CPh-m-CH 3 was used instead, three peaks corresponding to n = 17, 18, and 19 were found, which is consistent with the higher rened Au : Ag ratio in Au 19 Ag 27 (C^CPh-m-CH 3 ) 24 Cl 4 (PPh 3 ) 2 (Tables S3 and S4 †).The small inconsistency between MS and X-ray crystallography data has also been observed for other heterometallic NCs, which are known to be dynamic-and prone to rearrangement-in solution. 45,46egardless, the similar UV-vis and ESI-MS spectra for Au n -Ag 46−n (C^CPh) 24 Cl 4 (PPh 3 ) 2 and Au n Ag 46−n (C^CPh-m-X) 24 Cl 4 (PPh 3 ) 2 (X = F, CH 3 ) NCs-along with the fact that the same synthesis conditions were used-suggest that the three NCs are isostructural.Note that small differences in absorption features near 400 nm and shis in MS can be attributed to the different alkynyl ligands used.
The series of ve atomically precise Au/Ag NCs exhibit similar core structures.Adding or removing a single metal atom to or from the surface of NCs has provided insight into the optical and electronic properties of thiolate-capped metal NCs. 47,48The successful synthesis of our series of NCs with the same core (M 46 , M 44 , and M 43 ; M = Au and/or Ag) represents-to the best of our knowledge-the rst demonstration of atom-by-atom evolution for alkynyl-stabilized metal NCs structure (Fig. 2).Moreover, the linear directionality of alkynyl ligands in alkynylstabilized NCs offer advantages for catalytic studies since the NC/electrolyte interface is dominated by Au-C^C bonds with similar local arrangements.Note that for Au n Ag 46−n and Au 24 Ag 20 , Ag atoms are located in the inner dodecahedral Ag 12 shell and/or the icosahedral Au n Ag 12−n kernel which are not accessible to substrates interacting with the surface of the NC (Fig. 2, highlighted by the dashed red box).Importantly, the surface ligand density varies systematically across the series: 24 alkynyl, 4 chloride, and 2 phosphine ligands for Au n Ag 46−n , 24 alkynyl and 2 chloride ligands for Au 24 Ag 20 , and only 20 alkynyl ligands for Au 43 and Au 42 Ag 1 .Thus, this series of atomically precise NCs provides a powerful platform to study relationships between metal site accessibility and CO 2 RR activity.

Evaluation of CO 2 RR activity and selectivity
To investigate their efficacy for CO 2 RR catalysis, synthesized atomically precise NCs were mixed with carbon black (20 wt% NC loading) and deposited on a carbon paper electrode.For comparison, a carbon electrode was also prepared with spherical Au-SC 2 H 4 Ph NPs (Au-S NPs) with an average diameter of 3.1 ± 0.4 nm (Fig. S1C †) that were mixed with carbon black at the same mass loading.To provide an additional comparison, an 85 nm thick bulk gold layer was also deposited on one carbon paper electrode (referenced as Au layer) by electron beam deposition.Linear sweep voltammetry (LSV) was then performed for each electrode in CO 2 -saturated 0.5 M KHCO 3 solution.For the atomically precise NCs, the current density (j total ) was found to increase in the order of Au n Ag 46−n < Au 24 Ag 20 < Au 42 Ag 1 < Au 43 (Fig. 3A).Larger sized Au-S NPs exhibited a lower j total than all atomically precise NCs, and j total for the Au layer was the lowest of all catalysts investigated here (Fig. 3A).We note that the NC catalysts were not activated before electrocatalysis to minimize possible ligand stripping, and the LSV curves taken before and aer chronoamperometric CO 2 RR catalysis (potential range of -0.47 V to -0.77 V vs. RHE) were in close agreement (Fig. S7 †).Moreover, the NC-based catalysts display a steady current density over at least 40 min at each applied voltage (-0.47 V, -0.57V, -0.67 V and -0.77V vs. RHE; Fig. 3C), and the absorption spectra of NCs recovered from the electrode aer catalysis matched those of as-synthesized NCs (Fig. S8 †), conrming the stability of the NCs during CO 2 RR catalysis.
For all atomically precise NC catalysts evaluated here, CO was the major CO 2 RR product, and H 2 was the sole byproduct with no liquid products detected by 1 H NMR spectroscopy.The faradaic efficiency for CO production (FE CO ) was assessed for each catalyst by calculating the percentage of transferred charge that was directed toward CO production (see ESI † for details), and the highest FE CO were 92.1 ± 1.7% and 90.9 ± 1.4% for Au 43 and Au 42 Ag 1 , respectively, at a potential of −0.57 vs. RHE (Fig. 3B).At the same potential, FE CO for Au 24 Ag 20 and Au n -Ag 46−n were just 72.9 ± 1.0% and 59.5 ± 2.5%, respectively (Fig. 3B).These lower efficiencies can be attributed to more densely packed surface ligands, which likely favors the hydrogen evolution reaction (HER) over CO 2 RR. 25,26,49The Au-S NPs have an even lower FE CO (38.7 ± 0.8%) at the same potential, which is consistent with previously reported studies. 27,33If FE CO is related to the density of surface ligands, one might assume that a ligand-free Au layer would have the Fig. 4 The metal sites accessible to CO 2 on the surface of different Au/Ag-alkynyl NCs.
Table 1 The calculated number of accessible metal sites (N) for a series of alkynyl-protected NCs, along with their experimentally determined FE CO , j CO , TOF CO , and TOF CO /N values for CO 2 RR electrocatalysis at −0.57V vs. RHE.All experiments were conducted in a 0. highest efficiency.However, the FE CO for the Au layers is only 20.4 ± 1.6% at -0.57V vs. RHE, highlighting the important role of the microenvironments created by nanostructured catalysts in driving CO 2 RR. 3,50,51o further evaluate catalytic performance, CO partial current densities (j CO ) were determined and compared across the NC series.Generally, j CO decreases with increasing ligand density on the surface (Fig. 3D): Au 43 /Ag 42 Ag 1 (20 surface ligands) > Au 24 Ag 20 (26 surface ligands) > Au n Ag 46−n (30 surface ligands).For Au 43 , j CO is slightly higher than for Au 42 Ag 1 , suggesting that substitution of a single surface Au atom for Ag leads to a small decrease in CO 2 RR performance, particularly at more negative potentials.Furthermore, Au 43 shows a high CO turnover frequency (TOF CO ) of 4718 h −1 at −0.57V and 15 193 h −1 at −0.77 V vs. RHE (Fig. 3E), which exceeds the values for Au 24 Ag 20 (2597 h −1 at −0.57V and 10 658 h −1 at −0.77 V) and Au n Ag 46−n (1427 h −1 at −0.57V and 6400 h −1 at −0.77 V).Note that the same mass loading of NCs was used for all experiments.This makes it reasonable to directly compare TOF values since the NCs in this series have similar molecular weights (Table S5 †).In addition, differences in catalytic activity cannot be attributed to differences in NC stability as the FE CO of all catalysts remained constant for at least 2.5 hours at −0.57V (Fig. 3F).
While the nature of active sites is regarded as one of the best predictors of catalytic activity, 52 it is oen challenging to experimentally determine the number of catalytically active sites in a nanostructured material, and theoretical models are required. 53,54Though double-layer capacitance measurements can be used to determine the electrochemically active surface area (ECSA) of catalysts, the ECSA might not reect the surface area active specically for CO 2 RR since CO 2 RR and HER frequently occur simultaneously.For example, previous studies have shown that Au 25 (SR) 18 and Au 38 (SR) 24 exhibit different CO 2 RR behavior even though the NCs have almost the same ECSA. 33,49The ligand-to-metal ratio can serve as a proxy for active site density when the sizes of NCs are similar, 55 but this ratio is not directly related to the number of active sites owing to the different shapes and surface structures that similarly sized NCs can adopt.Indeed, two isomeric Au 38 (SR) 24 NCs with the same ligand-to-metal ratio have shown signicant differences in CO 2 RR catalysis. 34ith these challenges in mind, we sought to establish a simple method for determining the number of metal sites accessible to CO 2 in atomically precise NCs that does not rely on computationally intensive density functional theory (DFT) calculations.Briey, the NC structure determined by crystallography is used to generate a series of several thousand conformers accounting for the different ligand conformations that may arise due to ligand rotation in the absence of crystal packing effects (see ESI † for details).The accessible surface area of each conformation within the conformer series was then calculated for every surface atom using a 1.65 Å spherical probe (the kinetic radius of CO 2 , Fig. S9 †). 56Metal atoms with a positive contact area with CO 2 (Fig. S10 †) were counted as accessible since these atoms have sufficient space to accommodate a covalent bond with CO 2 .Accessible metal atoms for the different alkynyl-protected NCs under investigation are highlighted in Fig. 4. Note that the NC conformation with the greatest number of accessible metal atoms (N) was used to represent the surface accessibility of each NC.Using this approach, the number of accessible metal atoms for each NC can be calculated in ∼1 hour.Notably, the number of accessible metal atoms, N, for Au 43 /Au 42 Ag 1 , Au 24 Ag 20 , and Au n Ag 46−n are 16, 12, and 5, respectively, which is consistent with a greater density of CO 2 -accessible surface metal sites driving increased CO 2 RR activity.The role of CO 2 -accessible metal sites is further supported by the fact that we observe little variation in TOF CO when it is normalized to the number of accessible metal sites on each NC (TOF CO /N).Specically, TOF CO /N for Au 43 , Au 42 Ag 1 , Au 24 Ag 20 , and Au n Ag 46−n is 294 h −1 , 279 h −1 , 216 h −1 and 285 h −1 , respectively, at -0.57V vs. RHE (Table 1 and Fig. S11 †).This suggests that the number of CO 2 -accessible metal sites-rather than the degree of Ag doping, 57 surface ligand functional groups, 58 or the electronic structure 59 of the cluster-is the primary driver of catalytic activity, at least for NCs with relatively similar structures and compositions.
To further investigate the generalizability of accessible metal site number as a predictor of CO 2 RR activity, we also evaluated a wider range of previously reported alkynyl-protected Au NCs: Au 23 (C^C t Bu) 15 (Au 23 ), 60 Au 36 (C^CPh) 24 (Au 36 ), and Au 44 (-C^CPh) 28 (Au 44 ) 61 (Fig. S12 †).Within this series, Au 44 has the highest number of CO 2 -accessible metal sites (N = 12) and the highest CO 2 RR activity, while Au 23 has the lowest number of CO 2 -accessible sites and the lowest CO 2 RR activity (Fig. 5).The activity of Au 44 , however, is much lower than that of Au 43 , which is consistent with the greater number of CO 2 -accessible metal sites (N = 16) for the latter NC (Table 1 and Fig. S13 †).The relationship between the number of accessible metal sites and FE CO , j CO is plotted in Fig. S14.† This highlights that even though larger sized Au-alkynyl NCs oen have increased CO 2 RR activities, just like their Au-thiolate counterparts, 33 the number of accessible surface metals tends to be more closely related to catalyst performance.Therefore, we conclude that the number of accessible metal sites provides a useful metric for evaluating the likelihood of CO 2 binding to a particular atomically precise NC and for predicting trends in catalytic activity.Since Au 43 and Au 42 Ag 1 NCs are isostructural, their slight difference in CO 2 RR activity could be due to replacing a surface Au atom with a more electropositive Ag atom. 35Though Au 24 Ag 20 and Au 44 NCs have the same number of metal atoms (44) as well as the number of accessible metals (12), Au 44 exhibits a slightly higher FE CO and lower j CO than Au 24 Ag 20 .Differences in the geometric and electronic structures of the NCs may inuence their CO 2 RR performance, but this is likely a less signicant effect than the number of accessible metals.For NCs of the same core structure, more valence electrons in the frontier molecular orbitals elevate the energy of the highest occupied molecular orbital (HOMO), thereby favoring electron transfer from the NC catalyst to the substrate and thus improving the CO 2 RR activity. 59The performance of Au 43 and Au 42 Ag 1 NCs could also be partially attributed to more facile electron transfer during electrocatalysis.However, since the electronic structure of the NCs is determined by the number of metals and ligands, the surface coverage is still of importance.Moreover, since calculating the number of accessible metal sites is straightforward, it can serve as a quick screening tool for identifying the most promising NCs for electrochemical catalysis that is complementary to advanced DFT calculations.

Conclusions
A series of alkynyl-stabilized NCs with similar sizes and core structures but different degrees of surface ligand coverage was used to provide insight into the effect of the number of accessible metal sites on electrochemical CO 2 RR activity.A simple computational method was developed to calculate the number of metal sites on the NCs that are accessible to CO 2 .The highest faradaic efficiencies for CO 2 RR were observed for Au 43 and Au 42 Ag 1 , which feature the largest number of accessible metal sites.When the TOF CO of the NC-based catalysts was normalized by the number of accessible sites, the differences between NCs were reduced.Collectively, these trends suggest that the number of substrate-accessible metal sites serves as a useful and generalizable predictor for evaluating the potential of atomically precise NCs for CO 2 RR.
Specically, Au n Ag 46−n has an icosahedral Au n Ag 12−n kernel (n = 4-7 with Au and Ag randomly distributed), a dodecahedral Ag 20 inner shell, and an icosahedral Au 12 outer shell with two additional Ag atoms on the surface (Fig. 2A).Each of the additional Ag atoms on the surface of Au n Ag 46−n is bonded to two chloride ligands, one Au atom in the outer shell, and one PPh 3 ligand.Four chloride ligands are necessary for Au n Ag 46−n to adopt a closed-shell superatomic

Fig. 3
Fig. 3 (A) LSV and (B) FE CO for CO 2 RR for the series of NC-based catalysts.(C) Chronoamperometry data for the Au 43 -based catalyst at different applied potentials.(D) j CO and (E) TOF CO for the NC-based catalysts at different applied potentials during CO 2 RR.(F) FE CO for the NC-based catalysts at -0.57V vs. RHE during CO 2 RR over an extended time period.All experiments were conducted in a 0.5 M KHCO 3 solution saturated with CO 2 .In panels A, B, D, E, and F, magenta represents Au 43 (C^C t Bu) 20 , purple represents Au 42 Ag 1 (C^C t Bu) 20 , indigo represents Au 24 -Ag 20 (C^CPh t Bu) 24 Cl 2 , blue represents Au n Ag 46−n (C^CPh) 24 Cl 4 (PPh 3 ) 2 , sky blue represents Au-SC 2 H 4 Ph NPs, and cyan represents the Au layer.