The core premise of this contribution is to validate the direct participation of plasmon-generated hot carriers in a photocatalytic process. Plasmon hot carriers undergo rapid relaxation, resulting in notable local heat generation. Choosing a catalytic process where hot carriers and heat do not mediate equally is crucial. In simpler terms, an endothermic reaction with a substantial enthalpy difference (ΔΗ) is required.
CO2 conversion to CO is an endothermic reaction with a ΔΗ = 41 kJ/mol (equivalent to a temperature of 4930 K or 0.425 eV). The temperature is significantly higher than one can generate at the plasmon surface. Still, plasmons' hot electrons have higher energies than the requirement since they inject readily into semiconductors with Schottky barriers higher than 1.0 eV,35 even if the hole is behind transferred simultaneously.36 Generally speaking, thermal conversion of CO2 to CO typically requires high temperature, high pressure and a catalyst. The exact temperature required for this reaction would depend on the specific process and the reaction conditions. Still, according to Hunt et al. the reaction only starts at temperatures above 873 K. This offers a strategy to circumvent the engagement of the thermal process by crafting a nanohybrid system featuring a catalytic site that decomposes well before the initiation of the thermal process.
The direct electrochemical CO2 reduction conversion to CO via sequential one-electron steps involves going through the high-energy radical anion intermediate (CO2•−), demanding potentials higher than − 1.656 V vs SCE at pH 7., Therefore, evidence CO2 reduction in organic solvents under unbiased conditions, would be ascribed to hot electrons. Wang et al. showed that the reaction could be catalysed at room temperature after excitation multiple localised surface plasmon modes of aluminium nanoparticles with deep-UV range. Hence, by creating a catalytic site characterized by low thermal stability and substantiating the generation of CO in the absence of electrical bias in organic electrolytes under visible light excitation, it becomes feasible to affirm the engagement of plasmon hot electrons in the photocatalytic process.
The nanohybrid system was manufactured as a photoelectrode, schematically represented in Fig. 1. The photoelectrode consisted of a hole-accepting material (NiO), plasmonic material (Au nanoparticles (Au NPs)) and a catalyst (ReI(phen-NH2)(CO)3Cl), in a geometry consisting of fluorine-doped tin oxide (FTO) conductive glass with a NiO film deposited by screen printing, on which Au NPs were deposited via spray and later selectively functionalised by the Re catalyst through the -NH2 groups of the ligand. Transition-metal catalysts capable of facile multi-electron transformations offer an alternate pathway to circumvent this undesirable high-energy intermediate. Re, Mn, and Ru carbonyl catalysts containing diimine ligands based on 2,2′-bipyridine (bipy) have been investigated extensively for their CO2 reduction activity. The first catalyst of this family, the fac-Re(bipy)(CO)3Cl, was proposed in 1983. Since its discovery as an effective catalyst for CO2 reduction, rhenium(I) complexes have been the focus of intense studies.
FIGURE 2: (A) UV-Vis spectra of the thin films of Au, NiO, NiO-Au, and NiO/Au/ReI(phen-NH2)(CO)3Cl; (B) (i) FT-IR spectra of the ReI(phen-NH2)(CO)3Cl catalyst powder, (ii) FT-IR spectra of the NiO/Au/ReI(phen-NH2)(CO)3Cl thin film; (C) XPS (Al Kα source) and HAXPES (Ga Kα 9.2 keV source) of the NiO/Au/ReI(phen-NH2)(CO)3Cl as a film on FTO glass; and (D) Cyclic voltammogram of the ReI(phen-NH2)(CO)3Cl catalyst in acetonitrile solution and in N, N-dimethylformamide (DMF) solution using 0.1 M tetrabutylammonium hexafluorophosphate (TBAPF6) as supporting electrolyte, by using an Ag/AgCl (0.1 M TBAPF6/ acetonitrile) reference electrode, glassy carbon as working electrode, and Pt as a counter electrode.
As the catalyst, a mimic of the fac-Re(bipy)(CO)3Cl, was synthesizes with the difference being that the 2,2′-bipyridine ligand was replaced by the 1,10-Phenanthrolin-5-amine (phen-NH2). The terminal amine group permits direct coordination to the plasmonic gold surface. The attenuated total reflectance Fourier-transformed infrared (ATR-FTIR) spectrum of the complex is shown in Fig. 2B. In it is visible the carbonyl (1892 and 2019 cm− 1) and the primary amine stretching (3335 and 3434 cm− 1). The carbonyl peak at 1892 cm− 1 has double the intensity of the peak at 2019 cm− 1, suggesting that the two carbonyl groups are equivalents that is corroborated by the 13C NMR of the complex (Figure S3). The structure was further confirmed by proton nuclear magnetic resonance (1H-NMR) and corroborated by the optical spectrum since this complex class has characteristic absorptions below 400 nm. The complex's NMR (Fig. S4) and UV-Vis (Fig. S5) spectra can be found in SI.
Adding the catalyst to the NiO/Au film did not significantly change the part of the measurable optical spectrum because the substrate FTO glass has a sharp absorption edge at 360 nm, preventing us from detecting the absorption bands of the catalyst. ATR-FTIR confirmed the graphing of the catalyst to Au NPs at the first instance (Fig. 2B), where the amine groups' disappearance was noticeable due their bond to the Au NPs. The carbonyl region also changes in peak position and intensity ratio, not total intensity. In the grafted catalyst, none of the carbonyls is equivalent. The peaks are shifted to higher wavenumbers (1892, 1933 and 2029 cm− 1), suggesting electron donation from rhenium to the Au NPs breaking the original equivalency of two of the carbonyls. These observations confirm good electronic connectivity between the catalyst and plasmonic nanoparticles. The decrease in Au absorption peak intensity relates to increased film reflectivity and small heterogeneity across the film.
To further substantiate the anchoring of the molecular catalyst to the Au NPs, X-ray photoelectron spectroscopy (XPS) analyses were performed using two different excitation sources (Fig. 2C), namely Al Kα (1.48 keV) and Ga Kα (9.2 keV), the latter is commonly referred as hard X-ray photoelectron spectroscopy (HAXPES). The sources' energy differences enable different penetration probe depths. The analysis with Al Kα shows distinctly the peaks associated with Au and Re 4f together with the Ni 3p and O 2s. When the excitation energy source was increased (less surface sensitivity), the signal associated with Au and Re 4f reduced dramatically compared to Ni 3p and O 2s. The result suggests that Au and Re species are collocated as anticipated. The binding energies of Au and Re 4f7/2 were 84.0 eV and 41.9 eV, respectively, characterised by metallic Au and Re in the + 1 state. Ni 3p3/2 has 67.0 eV binding energy between Ni0 (65.9 eV) and Ni2+ (69.0 eV for NiO). The binding energy is close to what has been observed with Ni2+ in Ni2Ta (66.3 eV), suggesting a surface reduction of NiO surface. These states are responsible for the NiO visible absorption and are ascribed to trap states.
Figure 2D shows the catalyst's cyclic voltammetry (CV) profile in acetonitrile and DMF performed at 50 mV/s with 0.1 M TBAPF6 as a supporting electrolyte. It is clear that the solvents affect the reduction potenatial but not the shape of the CV, punctuated by two reduction peaks. The first reduction potential (-1.22 V vs Ag/AgCl in acetonitrile) is ascribed to ligand reduction, and the second at -1.75 V vs Ag/AgCl in acetonitrile is the reduction of the metal centre, consistent with previous studies.40 Further substantiation to this assignment will be provided when discussing ultrafast laser spectroscopy. For comparison purposes, the fac-Re(bipy)(CO)3Cl reduction peaks in 0.1 M TBAPF6 in acetonitrile are − 1.34 V and − 1.725 V vs SCE.52 The observed potential shift is consistent with ligand replacement. 52
The addition of the catalyst to the Au NPs (Fig. 3A) shifted the reduction peaks to -1.15 and − 2.05 V vs Ag/AgCl in acetonitrile, suggesting easier ligand reduction. At the same time, the reduction of the metal centre is more challenging, corroborating the ATR-FTIR that suggested electron donation from rhenium to Au NPs. When the Au NPs were supported on NiO, the process was reversed (-1.32 and − 1.88 V vs Ag/AgCl in acetonitrile), suggesting that NiO also donates electron density to Au NPs, thus competing with molecular catalyst donation. Upon adding CO2 to the solution, a catalytic wave was observed as soon as the metal is reduced (Fig. 3B insert and Fig. S6). The potential reaction mechanism will be discussed later on.
Chronoamperometry measurements during light on/light off measurements are shown in Fig. 3B at -1.8 V vs Ag/AgCl in acetonitrile. Noticeably, the photocurrent becomes detectable after − 1.7 V vs Ag/AgCl in acetonitrile. Using the integrated photocurrent and knowing the Re amount (0.18 wt%, determined by ICP-OES), it was possible to calculate the number of catalytic cycles per hour and the turnover frequency (TOF) at -1.8 V vs Ag/AgCl in acetonitrile. Under low CO2 solution saturation, the catalyst TOF was 0.156 min− 1. An average of ten catalytic cycles per hour due to illumination were recorded coupled to the low catalyst deactivation after 3h on stream is consistent with catalytic system not stoichiometric. The quantum yield was estimated to be 0.2%, assuming that each photon creates an electron-hole pair, which is not a perfect depiction of the plasmon process.
Examining catalytic performance parameters under constrained conditions concerning catalyst and CO2 amounts allows us to affirm the catalytic nature of the nanohybrid system. However, detecting photocurrent and, more importantly, product formation via quadrupole mass spectrometry (QMS) proved challenging. Consequently, comparable experiments were conducted, employing 1.5 times more catalyst on Au NPs and a heightened CO2 flow (20 mL/min). Before delving into the catalytic behaviour under the new conditions, it is noteworthy that the catalyst amount was kept deliberately low to prevent charge transfer between catalyst molecules. Furthermore, the amount of CO2 one can flow is capped at 20 mL/min so one can utilize the most sensitive detector in the QMS, and all the exhaust gas can be analyzed. The new conditions resulted in a six-fold the measured photocurrent (Fig. 3C), and one was able to detect a decrease in CO2 (m/z 44)/CO (m/z 28) ratio in the QMS when light was on (Fig. 3D) from 8.64 to 7.35, indicating formation of CO. Note that CO2 fragments into CO by about 8–10% depending on the QMS settings. Therefore, it is not prudent to report amount of CO form, instead one should present the ratio. Finally, it is worth mentioning that the experiments were performed in dry acetonitrile, which limits catalytic performance since protons are needed to extract the oxygen from the CO2 molecule. However, it reduced the chance for hydrogen evolution, a potential undesirable by-product, as shown in Fig. 3D.
The impact of illumination was most evident in the generated current rather than the overpotential, with only approximately a 50 mV shift detected. Consequently, it prompts a reasonable inquiry into whether these changes stem from hot electrons or the heat generated through plasmon thermalization. Additional support for hot electron involvement can be drawn from catalytic data. Firstly, the average temperature in the solution exhibited minimal change, staying within a 20 K range even under vigorous stirring. Secondly, the catalyst decomposed below 578 K, constraining the usable thermal window. Thirdly, the current exhibited an instantaneous response to light, aligning with expectations if hot carriers were involved. Lastly, the photosystem lacking NiO demonstrated inferior performance compared to the complete system, contradicting expectations if heat were the sole factor for enhanced catalysis. This discrepancy arises because NiO's role is to extract holes, thereby reducing recombination between electrons and holes, which would otherwise result in heat formation. Nevertheless, establishing more robust supporting evidence for hot carrier involvement is crucial. To achieve this, a series of impartial in situ studies were conducted. As the overall process involves catalytic reduction, providing evidence of the nanohybrid system's reduction without external potential would substantiate the role of hot electrons.
Unbiased ultrafast transient absorption spectroscopies were employed to scrutinize the participation of hot electrons in catalysis. Transient absorption spectroscopy (TAS) facilitates the monitoring of plasmon resonance (Fig. 4A). Analyzing the kinetics of the rapid signal decay allows for the extraction of the electron-phonon (e-ph) lifetime, a parameter highly sensitive to the number of carriers in resonance.34 The e-ph for Au nanoparticles on glass was estimated to be 8.5 ± 1.2 ps.
The e-ph process starts after the electron-electron (e-e) scattering process is found to happen less than 100 fs (shorter than the instrument response function), calculated from the TAS rising edge function that, and as expected., Link and El-Sayed showed that e-ph of Au nanoparticles around 15 nm in size in solution had an e-ph of 3–4 ps but increases with increased laser fluence and changes in a dielectric medium., Since the measurements were performed on low-density Au NPs films to avoid intraparticle interference, the laser fluency was on the high end of what Link & El-Sayed used to ensure good signal-to-noise data. Moreover, the samples were measured on solid substrates and, consequently, a different dielectric medium than in solution, furthering the increase in e-ph lifetime. Furthermore, the Au NPs were excited at 550 nm on the red side of the plasmon absorption peak to ensure that only the plasmon is excited, i.e., avoid intraband excitation. Considering all these factors, the estimated e-ph is within reason of what has been observed. Critically, the TAS of all the samples were performed under the same laser fluency and dielectric medium to ensure that the extracted e-ph lifetimes could be compared.
Depositing Au on NiO resulted in a decrease of the Au plasmon e-ph to 5.1 ± 0.6 ps. This is consistent with hole transfer from Au to NiO since the process increases the number of electrons in the resonance, which lowers the average temperature of the electrons in the resonance.34 Similarly, adding the catalyst to Au NPs also reduced the Au plasmon e-ph lifetime to 6.1 ± 0.4 ps. Plasmon resonance relates to particle morphology. However, the dynamic changes in the resonance are connected to the number of electrons participating in the resonance and their energy (or electron temperature). An electron acceptor (like the catalyst) reduces the resonance lifetime by taking hot electrons from it. Therefore, the observed change is a combination of two factors that affect the resonance in an antagonist way, namely, i) taking electrons from the resonance should increase the e-ph lifetime since fewer electrons share resonance energy, but ii) the electrons are removed hot (with higher kinetic energy) that reduces the average electron temperature that remains in the resonance and consequently e-ph lifetime.28 The complete system (NiO/Au/ReI(phen-NH2)(CO)3Cl) has an e-ph lifetime of 7.2 ± 0.6 ps, which is consistent with a reduction of e-ph due to energy (to the catalyst) and hole (to NiO) transfer that is slightly counteracted by the electron transfer to the catalyst.
To further substantiate the claim that holes and electrons are transferred to specific acceptors, unbiased transient infrared absorption spectroscopy (TIRAS) measurements were performed. TIRAS is highly sensitive to free carriers and changes in vibrations. TIRAS of Au on NiO revealed a broad featureless absorption (similar to the TIRAS contour plot of the complete system in Fig. 4B), characteristic of forming a quasi-metallic state due to free carriers.29 The hole injection time from the rising edge was 196 ± 91 fs. The signal decay was fitted with two exponential components, with the first component (τ1 = 0.45 ± 0.2 ps (> 90%)) accounting for most of the signal, making it difficult to determine τ2 reliably. More importantly, the signal decays completely in less than 1 ns, which means that under continuous illumination, it would be difficult to have charge accumulation necessary to catalyse reactions. The decays are associated with charge recombination at the interface of Au and NiO, similar to NiO-molecular dye hybrid systems. The shorter recombination time relates to charge recombination straight after injection, while the longer is associated with the charge that escaped from the interface to the bulk of NiO.
The complete system saw an even faster injection time of 124 ± 16 fs, as well as a slight increase in carriers' lifetime (Fig. 4C). The shorter-lived carriers required a fitting with two components τ1 = 0.25 ± 0.05 ps (76%) and τ2 = 2.57 ± 0.61 ps (20%), indicative that the addition of the catalyst prolongs the lifetime of the holes transferred to NiO since the catalyst extracts the electrons, effectively restoring Au NPs neutrality. Moreover, Fig. 4C reveals that 4% of the charges survive more than 1 ns, making them available for chemical reactions. Note that the TIRAS dynamics are without applied potential that will further expedite the charge-separated state lifetime and thus increase the amount of charge available for reaction.
One reason for emulating Lehn's catalyst is the remarkable sensitivity of the infrared-active carbonyl (C-O) stretching to electronic variations at the metal site. This characteristic renders it a perfect tool for examining alterations in the metal oxidation state. The transformation of CO2 to CO using a Re-carbonyl catalyst involves a two-electron stepwise reduction, with the initial reduction targeting the ligand. Consequently, notable alterations in the carbonyl stretching region should not be anticipated in the unbiased ultrafast TIRAS since only the initial reduction step is expected to be discernible at this rapid time scale.55 The complete systems should no detectable changes in the Re-carbonyl region (Fig. S7) except for the broad featureless signal related to free carriers, suggesting that the first electron transfer is to the ligand.
To further substantiate this hypothesis, the TIRAS of the Au-Re catalyst without NiO was measured (Fig. 4D). The idea behind this experiment was to reduce signal overlap with hot hole in NiO. Despite the low signal-to-noise, the system Au-Re catalyst excited to the red of the maximum plasmon absorption (to avoid intraband excitation) yielded a broad and featureless signal with positive absorption. Since there is no semiconductor and the plasmon excitation energy prevented intraband excitation, the signal must relate to the injected electron in the catalyst. Based on the CV data and the absence of changes in the Re-carbonyl region, one can establish that the signal relates to electron reduction of the phenanthroline ligand. The signal shape suggests that the electron is delocalised through the aromatic rings, acting like a free carrier.
To further support this hypothesis, the TIRAS of the Au-Re catalyst without NiO was conducted (refer to Fig. 4D). The rationale behind this experiment was to minimize signal interference from hot holes in NiO. Despite the challenge of low signal-to-noise, when the Au-Re catalyst was excited to the red of the maximum plasmon absorption (to avoid intraband excitation), it produced a broad and featureless signal with positive absorption. Given the absence of a semiconductor and the fact that plasmon excitation energy prevented intraband excitation, it can be concluded that the signal is linked to the injected electron in the catalyst. By analyzing CV data and observing no changes in the Re-carbonyl region, it can be inferred that the signal corresponds to the electron reduction of the phenanthroline ligand. The signal's shape implies that the electron is distributed across the aromatic rings, behaving akin to a free carrier.
Having established that plasmon-induced hot electrons reduced the linker through ultrafast spectroscopy, the subsequent step involved substantiating the second reduction leading to CO formation. This verification was done through in situ unbiased Fourier-transformed infrared (FT-IR) spectroscopy experiments conducted under 532 nm illumination. These experiments were executed in acetonitrile saturated with CO2 and argon.
Figure 5A shows the temporal evolution of the FTIR signal outside the region of interest (carbonyl region 2100 − 1900 cm− 1), which reveals a background shift with increased illumination. A similar transient behavior was detected in argon but not visible when the system was not irradiated, confirming that this signal is related to the effect of light not reacting to the atmosphere. This light-dependent background shift is ascribed to free carriers,69, in the NiO and ligand, corroborating the ultrafast transient data.
The background shift due to free carriers makes data analysis challenging. The change was correct, fitting a baseline with the same parameters. By subtracting the spectra from the spectrum collected in the dark, it enabled analysis of the temporal evolution of the carbonyl groups not affected by the acetonitrile absorption. Unfortunately, acetonitrile has strong absorption bands between 1800–1900 cm− 1, making it challenging to follow the equivalent carbonyl peaks. Therefore, the analysis was restricted to the inequivalent carbonyl centered at 2024 cm− 1 (ATR-FTIR 2019 cm− 1). Solvation effects can justify the slight difference.
It is evident that under illumination, the 2024 cm− 1 peak decreases in intensity and a new peak appears at 2005 cm− 1. The reduction in the band centered at 2024 cm− 1 related to the inequivalent carbonyl is correlated with ligand reduction because this was, to some extent, observed in the sample illuminated in argon. The ligand reduction appears to induce a structural change in the catalyst complex, making all the CO originally coordinated to the Re+ centre equivalent. The possibility of ligand detachment was discarded because one did not free CO, due to the absence of its characteristic band at 2193 cm− 1 in the spectra, and the dry catalyst FTIR was the same before and after illumination in argon. Klein et al. proposed a similar in the catalyst with a bipy ligand. According to them, the ligand plays a critical “noninnocent” role by storing the first reducing equivalent in a ReI(bipy•−) state, leading to a complex reorganisation. The appearance of the peak at 2005 cm− 1 seems consistent with the formation of a new ReI-CO species since the finding was only visible when the sample was illuminated and was in the presence of CO2, consistent with what Rotundo et al.71 proposed. Their mechanism suggests the formation of such species after two proton additions and the liberation of H2O. Both observations suggest catalyst reduction and the appearance of CO without applied potential, consistent with a mechanism involving hot carriers.
Considering the catalytic and characterization data, we proposed a catalytic mechanism depicted in Fig. 6. The initial step is the coordination of CO2 to the catalyst by exchange with the Cl− group, consistent with previously reported., Excitation of Au NPs LSPR forms hot electrons and holes. The hot holes are injected into the NiO and reacted on the counter electrode, producing O2. The first electron reduces the ligand consistent with the TIRAS measurements and the most updated mechanism proposals for Lehn’s catalyst.71 After this, the bonded CO2 molecule undergoes two sequential protonations, forming a new CO species bonded to Re+ (confirmed by unbiased in situ FTIR) and the liberation of one H2O molecule. The second protonation, according to Rotundo et al.,71 is slightly endothermic, explaining at least partially the fact that the newly formed ReI-CO species FTIR band is not very intense after 30 min on stream. This is further supported by an increase in the CV catalytic wave current when a more potent proton donor (i.e. 1,1,1,3,3,3-Hexafluoro-2-propanol (HFIP), was introduced (Figure S8). The CO molecule is released with the second reduction by the plasmon hot electrons, regenerating the catalyst and enabling the binding of a new CO2 molecule.