## Introduction

Carbon dioxide (CO2) fixation is recognized to be a much-needed component of a carbon-neutral energy strategy1,2,3,4. Although CO2 is relatively unreactive, various catalytic processes triggered by heat (thermochemical)5,6,7,8, electricity (electrochemical)9,10,11,12,13,14,15,16,17, and light (photochemical)18,19,20,21,22,23,24,25,26,27,28 are being explored for activating CO2 and recycling it back to valuable petrochemicals. Sunlight-driven conversion of CO2 to fuels is particularly attractive as a means to store intermittent solar energy in the form of C–C and C–H bonds. Semiconductor and metal-catalyzed photoelectrolytic reduction of CO2 has shown promise; however, these processes have often required ultraviolet (UV) light and/or considerable electrical energy input, or they do not favor energy-rich hydrocarbon products. Longer-chain hydrocarbons possess higher energy densities. Moreover, hydrocarbons in the liquid state are easier to transport29,30. However, the formation of longer-chain hydrocarbons from CO2 requires multiple electron (e) and proton (H+) transfer steps, as well as C–C bond formation9,31,32, which pose major kinetic bottlenecks.

Here we demonstrate a visible-light-driven route for the conversion of CO2 and H2O into C1–C3 hydrocarbons. The scheme does not involve the application of an electrochemical potential, UV light, high temperatures, hydrogen gas, or a sacrificial agent. It uses green light as the sole energy input and driving agent. The strategy employs plasmonic Au nanoparticles (NPs) of a pseudospherical shape and an average diameter of ~12 nm, as characterized previously28. Au NPs are known from electrochemical studies33 to activate CO2. The choice of Au NPs was further driven by the relative chemical stability of Au against bulk oxidation and photocorrosion; the other two common plasmonic metals, Ag and Cu, while electrocatalytically active for CO2 reduction, are prone to oxidation in air, water, and/or light excitation. The Au NPs possess a strong localized surface plasmon resonance (LSPR) band centered around 520 nm (Fig. 1a), which enables strong, resonant absorption of green light. The LSPR excitation of the NPs yields energetic electron–hole (e–h+) carriers via Landau damping. These e–h+ carriers were shown in recent studies to drive redox conversions28,34,35,36, especially the conversion of CO2 to methane and ethane under blue–green light28. However, in this past demonstration, isopropanol was used as a sacrificial h+ scavenger to facilitate e–h+ pair separation; otherwise, unproductive e–h+ recombination dominated. Thus, isopropanol served as the H+ source in this CO2 reduction scheme, which posed a major limitation for net energy storage.

The present strategy overcomes this drawback and uses water as the H+ source and does not require a sacrificial h+ scavenger, thus constituting a truly fuel-forming reaction. The enhanced reactivity was enabled by the use of an ionic liquid (IL) medium, specifically comprised 1-ethyl-3-methylimidazolium tetrafluoroborate (EMIM-BF4). Our choice was motivated by examples from electrocatalytic CO2 reduction reaction (CO2RR) where the EMIM-BF4 electrolyte, owing to its highly ionic character, stabilizes the high-energy CO2 radical anion intermediate formed in the reaction and decreases the overpotential needed for CO2RR37,38,39,40,41,42. In addition, EMIM-BF4 has a wide electrochemical window and high thermal stability43,44. In our photocatalytic scheme, the EMIM-BF4, as we find from kinetic analysis and density functional theory (DFT) simulations, promotes e transfer at the interface of the photoexcited Au NP and adsorbed CO2 (Fig. 1b), obviating the need for a h+ scavenger or applied potential for e–h+ separation.

## Results

### IL-mediated plasmonic CO2 reduction

The photocatalyst had the form of a substrate-supported film of Au NPs immersed in an aqueous solution of EMIM-BF4 saturated with CO2 and contained inside a glass reactor (Supplementary Methods). The light excitation source comprised a continuous-wave (CW) laser of a wavelength of 532 nm light and an intensity of 1 W cm−2. Under CW excitation, the steady-state temperature of the reaction medium got moderately elevated to ~48 °C. Hydrocarbon products collected in the reactor headspace were measured (Supplementary Figs. 111) using a gas chromatograph (GC) equipped with a flame ionization detector. The EMIM-BF4 concentration was varied from 0 to 100 mol%, to find optimal conditions for CO2RR. In 1–10 mol% EMIM-BF4, the products of plasmon-excitation-driven CO2RR were found to be C1 (CH4), C2 (C2H4 and C2H2), and highly reduced C3 (C3H6 and C3H8) hydrocarbons (Fig. 1c, d and Supplementary Note 1). This product profile is quite striking when one considers that the major product in electrochemical CO2RR is carbon monoxide (CO) formed by 2e–2H+ reduction of CO2 (refs. 13,14,15,16,17). On the other hand, propane (C3H8), formed in our scheme, requires an overall 20e–20 H+ reduction and coupling of three CO2 molecules. Such generation of C3 hydrocarbons by artificial photosynthesis is challenging and therefore rare.

The CO2RR activity depends on the IL concentration (Fig. 1c). In pure water the activity was nil, whereas in 1 mol% EMIM-BF4 solution the generation of C1, C2, and C3 hydrocarbons was observed. The CO2RR activity, as quantified by turnover frequencies (TOFs) of the hydrocarbon products, increased dramatically with an increase in the EMIM-BF4 concentration. The highest activity was found at 5 mol% EMIM-BF4. Increasing the EMIM-BF4 concentration further resulted in a sharp drop in the CO2RR activity. In 100 mol% EMIM-BF4 solution, the activity was nil, similar to that in pure water. Thus, the CO2RR activity exhibits a volcano relationship as a function of the EMIM-BF4 concentration (Fig. 1c). At all EMIM-BF4 concentrations, where C1, C2, and C3 hydrocarbons were produced, the product selectivity was found to follow the order: C1 > C2 > C3. The selectivity for C2+ production is ~50% in 1−10 mol% EMIM-BF4 solution (Fig. 1d).

Non-hydrocarbon products were also characterized by a GC equipped with a thermal conductivity detector (TCD) (Supplementary Figs. 1215). Considerable hydrogen (H2) production was measured (Supplementary Fig. 12), the TOF of which was 138.2 NP1 h−1 in 5 mol% EMIM-BF4 solution, the IL concentration where CO2RR activity is the highest. The H2 likely originates from the competing reduction of H+ in the reaction medium (Supplementary Eq. (6)). In the GC-TCD measurements, there were no detection of CO (Supplementary Fig. 15), otherwise known to be a major product in electrocatalytic CO2RR on Au (refs. 13,14,15,16,17). Of the possible oxidation products, there was no measurable production of O2 (see Supplementary Information). H2O2 was detected (Supplementary Figs. 1618) by the fluorogenic test employing a amplex red and horseradish peroxidase reagent45. Thus, the oxidation of H2O to H2O2 and H+ (2H2O → H2O2 + 2H+ + 2e) is the likely oxidation half-reaction that consumes the photogenerated h+.

Control studies were performed, one without Au NPs, another without light, and a third without CO2. The conditions were otherwise maintained the same as those in the photoreaction tests and a 5 mol% EMIM-BF4 solution, found to be most optimal in the photoreaction tests, was employed. The control studies showed that the absence of any one of the components Au NPs, green light illumination, or CO2 resulted in nil hydrocarbon production, despite the use of 5 mol% EMIM-BF4 solution (Supplementary Fig. 19a–c). Thus, it is confirmed that the hydrocarbon production originates from green-light-driven CO2 reduction on Au NPs. The control study without light excitation was performed at an elevated temperature of 50 °C so as to mimic the steady-state bulk solution temperature of the reaction mixture in the photoreaction tests. The lack of CO2RR activity in this dark control study demonstrates that the CO2RR activity in the photoreaction tests does not originate from simply a photothermal effect of the light excitation. Rather a photoredox process facilitated by the Au NPs and the IL is responsible for the conversion of CO2 to hydrocarbons.

The plasmonic catalyst also exhibited stability and recyclability under the photoreaction conditions and IL media subjected on the catalyst. We tested the same substrate-supported Au NP film immersed in 5 mol% EMIM-BF4 over multiple cycles, each consisting of a 10 h photoreaction. The CO2RR activity and product selectivity, as determined from the TOFs of the hydrocarbon products, was maintained over the course of this multi-cycle test (Supplementary Fig. 20). As the NP film or EMIM-BF4 solution were not replenished between cycles, the maintenance of CO2RR activity over multiple cycles suggests that Au and EMIM-BF4 were not consumed, at any discernible levels, in the photoredox reaction.

### The origin of products

Given the hydrocarbon profile of the product mixture, it was necessary to go beyond the control studies described above and confirm more directly that CO2, rather than carbon contamination or photolysis of the EMIM-BF4, was the source of the hydrocarbon products. For this confirmation, 13C isotope labeling was employed (Fig. 2 and Supplementary Figs. 21 and 22). In this labeling study, 13CO2 was employed as the reactant instead of 12CO2, whereas all other conditions were kept the same as those in other photoreaction tests. GC-mass spectrometry (GC-MS) was used for identification of the hydrocarbon products generated in the photoreaction (Fig. 2a). The GC-MS analysis confirmed the presence of 13CH4 (Fig. 2b) and 13C2H2 (Fig. 2c), manifested by their characteristic mass fragmentation patterns, shifted to higher m/z compared with reference fragmentation patterns of 12CH4 and 12C2H2, respectively. Thus, isotope labeling confirms CO2 to be the origin of hydrocarbon products.

### The role of the IL

We attempted to gain a mechanistic understanding of this catalytic scheme focusing on the question of how the IL promotes CO2RR activity. It was observed that the presence of EMIM-BF4 in the aqueous medium results in a considerably acidic pH (Supplementary Fig. 23): the 5 mol% EMIM-BF4 solution has a pH of 2.95. To determine whether this acidity is responsible for the enhanced CO2RR activity in a EMIM-BF4 solution, we performed a photoreaction in deionized water containing no EMIM-BF4 but with a pH of 2.93 achieved using acid (Supplementary Fig. 19d). All other conditions were kept the same as in the photoreactions in EMIM-BF4 solutions. In this EMIM-BF4-free photoreaction, no products were observed, which demonstrated that the high acidity or H+ concentration, [H+], of the EMIM-BF4-containing medium is not the sole cause of the enhanced CO2RR activity. EMIM-BF4 plays other role(s). It is possible, in principle, for EMIM-BF4, instead of H2O, to serve as the h+ acceptor; however, if this were the case, then the CO2RR activity would have been enhanced at higher EMIM-BF4 concentrations, in line with a study of a different plasmon excitation-catalyzed redox reaction36. Instead, we observed peak activity at a EMIM-BF4 concentration of 5 mol%, above which the activity drops steeply reaching nil in pure EMIM-BF4 wherein H2O is not available.

We hypothesized that the strongly ionic character of EMIM-BF4 plays a role in the activation of CO2, which is otherwise fairly redox inactive. CO2, however, is highly polarizable, as indicated by its quadrupole moment of −4.3 D Å (ref. 46). The interaction of EMIM-BF4 and CO2 was simulated by DFT. A past study suggests that CO2 can undergo complexation with the N-heterocyclic carbene, EMIM*, formed from EMIM+ by H+ loss42. We investigated using DFT the structure of such a [EMIM*-CO2] complex (Fig. 3a). The complex exhibits binding between the C atom of the CO2 and the C2 atom of the imidazole ring with an energy of intermolecular interaction, Em-m, of −0.36 eV. This interaction is stronger than, for instance, the interaction of an H2O molecule and CO2 (Fig. 3b). Unlike the latter case, complexation with EMIM* leads to considerable restructuring of the CO2 moiety. The CO2 moiety adopts a bent configuration with an O=C=O angle of 133.7° and C=O bonds lengthened to 1.24 Å. In fact, the geometry of the CO2 moiety in the complex closely mirrors that of the CO2•− anion radical, which has a bond angle of 137.8° and bond length of 1.23 Å (Supplementary Fig. 24). Moreover, from Mulliken charge partitioning analysis (Supplementary Fig. 25), the CO2 moiety in the [EMIM*-CO2] complex is found to have a net charge of −0.73, which indicates its partial anionic character.

It is known that the energetic cost of the drastic structural reorganization from linear CO2 to the bent CO2•− anion radical poses a major barrier for e acceptance by CO2 (refs. 37,38,39,40,41,42). However, our DFT calculations show that in its complex with EMIM*, the CO2 moiety is structurally pre-configured for e acceptance. Consistent with this finding, 1e addition to [EMIM*-CO2] is much more favorable as compared with 1e addition to CO2 (Fig. 3c, d). Thus, it appears that EMIM-BF4 can promote the transfer of photogenerated e from the Au NP to adsorbed CO2, which is otherwise a major kinetic bottleneck in the photocatalytic reduction process. Furthermore, it is plausible that the CO2•− anion radical formed on the Au surface by photo-initiated e transfer process has an enhanced lifetime due to solvation or complexation by EMIM+ (Fig. 3e). A longer lifetime of this reactive intermediate would increase the probability of C–C coupling between the intermediates.

### Empirical kinetic model

Although the DFT computations provide insight into the central role of EMIM-BF4 in CO2 activation, the volcano-type dependence of the CO2RR activity on the IL concentration deserves an explanation. From the hydrolysis of EMIM-BF4 known from past studies47,48,49,50:

$${\mathrm{EMIM-BF}}_4 + x{\mathrm{H}}_2{\mathrm{O}} \to {\mathrm{EMIM}}^ + + \left[ {{\mathrm{BF}}_{4-x}\left( {{\mathrm{OH}}} \right)_x} \right]^- + x{\mathrm{HF}}$$
(1)

where x = 1–4 and the complexation of CO2 with EMIM+ predicted in DFT simulations:

$${\mathrm{EMIM}}^ + + {\mathrm{CO}}_2 \to \left[ {{\mathrm{EMIM}}^ \ast{\mathrm{-CO}}_2} \right] + {\mathrm{H}}^ +$$
(2)

we postulate a rate determining step in the reaction of CO2 and H2O:

$${\mathrm{EMIM-BF}}_4 + {\mathrm{CO}}_2 + x{\mathrm{H}}_2{\mathrm{O}} \to \left[ {{\mathrm{EMIM}}^\ast{\mathrm{-CO}}_2} \right] \\ + \left[ {{\mathrm{BF}}_{4-x}\left( {{\mathrm{OH}}} \right)_x} \right]^- + \left( {x + 1} \right){\mathrm{H}}^ + + x{\mathrm{F}}^-$$
(3)

From this reaction equation, the concentration of the activated CO2 complex, [EMIM*-CO2], is expected to be directly proportional to [H+]x+1. Therefore, the [H+] determined from the measured pH of the EMIM-BF4 solution (Supplementary Fig. 23) serves as a proxy for the concentration of [EMIM*-CO2], based on which the [EMIM*-CO2] concentration is expected to be the highest in the EMIM-BF4 concentration range around 5 mol%. The higher the concentration of the activated [EMIM*-CO2] complex, the greater is the rate of CO2 conversion and also the higher the likelihood of C–C coupling required for C2+ production. Therefore, both the overall activity and the selectivity in favor of C2+ products are favorable in the 3–7 mol% EMIM-BF4 range, with the most optimal performance achieved at 5 mol% EMIM-BF4. On the other hand, the activated complex has zero concentration in pure water on one extreme and in pure EMIM-BF4 on the other extreme, which explains the nil turnover at these conditions. An additional reason for the drop in activity at higher EMIM-BF4 concentrations may be that the adsorption of BF4 to the Au NP surface (Supplementary Fig. 26) dominates at these concentrations to such an extent that the adsorption of CO2 and/or [EMIM*-CO2] to the Au surface is largely inhibited and so is the e transfer to CO2.

The CO2RR activity depends on the concentration of this activated complex to a high reaction order. This is best exemplified by the plots of TOF for each hydrocarbon as a function of the [H+] (Fig. 4a–e), which as explained above, serves as a proxy for the concentration of [EMIM*-CO2]. The pseudo-reaction order, n, is found to be 1.9 for C2H4, 2.5 for C2H2, 3.7 for C3H6, and 4.0 for C3H8. The fit for the CH4 TOF has a relatively high χ2-value, so the n of 2.7 estimated for CH4 has a lower confidence. In general, the pseudo-reaction order is higher for the longer hydrocarbons, which perhaps captures the need for multiple activated complexes to be available for undergoing coupling to C2 and C3 fragments. The high pseudo-reaction order for the C3 products goes hand-in-hand with an apparent threshold in [H+] below which the TOF is zero or below the detection limit (Fig. 4d, e). For each of the hydrocarbon products, the [H+] raised to the power of the corresponding n follows a volcano trend with respect to the EMIM-BF4 concentration, mirroring closely the trend in the TOF for that hydrocarbon (Fig. 4f–j).

Thus, we reported the green-light-driven synthesis of C1–C3 hydrocarbons from CO2 and water on plasmonic Au NPs in an IL medium. The resonant green light absorption of the plasmonic NPs and their ability to sustain electrostatically charged surfaces under resonant CW excitation are at the heart of the observed photoreactivity. The IL plays a synergistic role due to its complexation with the CO2, which preconfigures the CO2 for accepting e from photoexcited Au NPs. The enhanced reactivity of CO2 in the presence of the IL obviates the need for an applied potential or a sacrificial scavenger. Although hydrocarbon production yields in the reaction need further optimization, the generation of propane by overall 20e–20H+ reduction and coupling of three CO2 molecules is both striking and mechanistically rich. The precise intermediates and reaction pathways, including C–C coupling and dehydrogenation steps, which yield each of the hydrocarbons, deserve further elucidation. Beyond CO2 conversion studied here, ILs may have promise in other photocatalytic schemes where activation of relatively inert substrates and stabilization of high-energy charged intermediates is desirable.